Mathematical Modeling for Prediction of Survival From Resuscitation Based on Computerized Continuous Capnography: Proof of Concept

Authors

  • Sharon Einav MD,

    1. From the Adult (SE) and Neonatal (RB) Critical Care Units, Shaare Zedek Medical Center, and the Department of Anesthesia (CFW), Hadassah-Hebrew University Medical Center, Jerusalem; and the Department of Anesthesia of the Sourasky Medical Center (IM), Tel-Aviv, Israel.
    Search for more papers by this author
  • Ruben Bromiker MD,

    1. From the Adult (SE) and Neonatal (RB) Critical Care Units, Shaare Zedek Medical Center, and the Department of Anesthesia (CFW), Hadassah-Hebrew University Medical Center, Jerusalem; and the Department of Anesthesia of the Sourasky Medical Center (IM), Tel-Aviv, Israel.
    Search for more papers by this author
  • Carolyn F. Weiniger MB, ChB,

    1. From the Adult (SE) and Neonatal (RB) Critical Care Units, Shaare Zedek Medical Center, and the Department of Anesthesia (CFW), Hadassah-Hebrew University Medical Center, Jerusalem; and the Department of Anesthesia of the Sourasky Medical Center (IM), Tel-Aviv, Israel.
    Search for more papers by this author
  • Idit Matot MD

    1. From the Adult (SE) and Neonatal (RB) Critical Care Units, Shaare Zedek Medical Center, and the Department of Anesthesia (CFW), Hadassah-Hebrew University Medical Center, Jerusalem; and the Department of Anesthesia of the Sourasky Medical Center (IM), Tel-Aviv, Israel.
    Search for more papers by this author

  • The preliminary results of this study were presented at the 9th Scientific Congress of the European Resuscitation Council, Ghent, Belgium, 2008.

  • This study was supported by a grant from the Chief Scientist of the Ministry of Health, Jerusalem, Israel (Grant 5397). Oridion provided the capnography equipment, developed the dedicated software, and funded the mathematician who performed data extraction. Neither the Ministry of Health or Oridion had a role in study design, data analyses and interpretation, writing of the manuscript or manuscript submission for publication.

  • The authors have no relevant financial information or potential conflicts of interest to disclose.

  • Supervising Editor: James E. Olson, PhD.

Address for correspondence and reprints: Sharon Einav, MD; e-mail: einav_s@szmc.org.il.

Abstract

ACADEMIC EMERGENCY MEDICINE 2011; 18:468–475 © 2011 by the Society for Academic Emergency Medicine

Abstract

Objectives:  The objective was to describe a new method of studying correlations between real-time end tidal carbon dioxide (ETCO2) data and resuscitation outcomes.

Methods:  This was a prospective cohort study of 30 patients who underwent cardiopulmonary resuscitation (CPR) in a university hospital. Sidestream capnograph data were collected during CPR and analyzed by a mathematician blinded to patient outcome. The primary outcome measure was to determine whether a meaningful relationship could be drawn between detailed computerized ETCO2 characteristics and the return of spontaneous circulation (ROSC). Significance testing was performed for proof-of-concept purposes only.

Results:  Median patient age was 74 years (interquartile range [IQR] = 60–80 years; range = 16–92 years). Events were mostly witnessed (63%), with a median call-to-arrival time of 150 seconds (IQR = 105–255 seconds; range = 60–300 seconds). The incidence of ROSC was 57% (17 of 30), and of hospital discharge 20% (six of 30). Ten minutes after intubation, patients with ROSC had higher peak ETCO2 values (p = 0.035), larger areas under the ETCO2 curve (p = 0.016), and rising ETCO2 slopes versus flat or falling slopes (p = 0.016) when compared to patients without ROSC. Cumulative maxETCO2 > 20 mm Hg at all time points measured between 5 and 10 minutes postintubation best predicted ROSC (sensitivity = 0.88; specificity = 0.77; p < 0.001). Mathematical modeling targeted toward avoiding misdiagnosis of patients with recovery potential (fixed condition, false-negative rate = 0) demonstrated that cumulative maxETCO2 (at 5–10 minutes) > 25 mm Hg or a slope greater than 0 measured between 0 and 8 minutes correctly predicted patient outcome in 70% of cases within less than 10 minutes of intubation.

Conclusions:  This preliminary study suggests that computerized ETCO2 carries potential as a tool for early, real-time decision-making during some resuscitations.

Resuscitative efforts are unsuccessful in most patients undergoing cardiopulmonary resuscitation (CPR), and survivors often sustain severe neurologic injury leading eventually to death.1 In an attempt to curb effort and expenditure on futile CPR, some suggest that termination of CPR be considered in certain patients who are unresponsive to advanced life support efforts for more than 20 minutes.2,3

The potential value of capnography as a predictive tool in resuscitation has been explored in the past.4–17 Capnography has been integrated into several types of defibrillators that are currently in use by prehospital emergency medical services (EMS). Use of this tool, however, has been limited by the subjectivity of manual recording techniques (Figure 1) and the presence of outliers (i.e., patients who recovered despite predictions to the contrary).

Figure 1.

 Capnography trace of a patient during CPR. The arrow denotes the period during which chest compression was performed. Resuscitation efforts were discontinued following ROSC, but the patient gradually deteriorated hemodynamically. Note the large swings in ETCO2 during resuscitative efforts, largely due to changes in flow secondary to chest compression and decompression. Studies using manual ETCO2 monitoring for prediction of outcome are based on selection of a single value read on the monitor at a specified time point. The dotted line marks approximately 10 minutes into the resuscitation. The two circles show two very different values that could have been selected had the manual technique been used at this time. Computerized monitoring enables real-time calculation of values which better reflect the overall form of the ETCO2 trace over time. CPR = cardiopulmonary resuscitation; ETCO2 = end tidal carbon dioxide; ROSC = return of spontaneous circulation.

This study was designed to determine whether a meaningful relationship could be drawn between end tidal CO2 (ETCO2) and survival outcomes by detailed computer analysis and mathematical modeling of continuous ETCO2 data recorded during resuscitation. This approach differs from previous reports correlating ETCO2 with survival in the accuracy, detail, and amount of data incorporated into the correlation (and consequently into later decision-making) and is less biased than manual recording of data sets at preset time points. This was a pilot study, intended only to describe how characteristics of the ETCO2 data may inform the design of the ETCO2 analysis to generate a prediction model. The small sample of patients enrolled in this study limits the drawing of meaningful inferences regarding survival from the data gathered.

Methods

Study Design

This was a prospective, observational pilot study. The study protocol was reviewed by the institutional review board and granted a waiver of informed consent.

Study Setting and Population

The study was performed in a tertiary medical center with a 24-hour on-call, dedicated, certified, CPR team (composed of a cardiologist, an anesthesiologist, and a trained nurse) who are summoned via pager by a call from any hospital telephone. Resuscitation is performed with standardized equipment and according to American Heart Association guidelines. Intubation is performed by the anesthesiologist immediately upon arrival on scene. Decisions to terminate resuscitation efforts are made collectively by the physicians. The quality of resuscitation is later reviewed on a case-by-case basis by a dedicated committee.

Data from a convenience sample of patients were prospectively collected during a 6-month period. Included were patients who sustained in-hospital nontraumatic cardiac arrest requiring advanced life support during the morning shift. Patient selection was guided by the availability of the primary investigator. It was felt that due to the “uncontrolled” nature of an in-hospital emergency scenario, an external observer should be on location to verify that no protocol violations had occurred (e.g., inclusion/exclusion criteria, timing of capnograph connection). Excluded were patients younger than 16 years; those with do not attempt resuscitation orders, multiple intubation attempts, or capnographic measurements that were not initiated immediately following tracheal intubation; and patients whose resuscitation attempts lasted less than 15 minutes (due to the a priori requirement for a monitoring period of sufficient length to be used for analysis).

Study Protocol

Cardiac arrest was defined as the absence of effective spontaneous respiration and a palpable pulse determined by at least two CPR team members and electrocardiographic evidence of a nonperfusing rhythm (e.g., asystole, ventricular fibrillation/ventricular tachycardia [VF/VT]) when relevant. Return of spontaneous circulation (ROSC) was defined as the appearance of a palpable pulse and a measurable blood pressure lasting at least 1 hour subsequent to the termination of resuscitative efforts.

Event details were recorded on scene by the responding nurses (code team and ward). The data were then validated within 24 hours by a study investigator through debriefings with department medical staff and members of the code team who had attended the event. Full inpatient medical records were retrieved from patient charts within 24 hours of the event. Patient follow-up was continued until in-hospital death or, alternatively, to hospital discharge.

Capnographic Data

Acquisition.  A portable side-stream capnograph (Microcap Plus, Oridion, Jerusalem, Israel) was connected between the endotracheal tube and the manual resuscitation bag immediately following intubation. Sidestream capnographs continuously draw sample gases via an endotracheal tube adapter through a sampling tube to an infrared light source and detector within a remote monitor. The devices sample air at about 50 mL/min. Major limitations of this technique when compared to mainstream capnography (where the infrared source and detector are mounted lateral to the adaptor itself) include use in certain low gas flow applications and a slightly less accurate measurement. Inherent advantages include the remote mounting of the expensive optical sensors (not directly on the endotracheal tube adapter) making it less vulnerable to damage, a lower likelihood that readings are disrupted/obstructed by moisture condensation or respiratory tract secretions, and no increased bulk on the endotracheal tubing, making it less prone to kinking. In the current study, ETCO2 data were recorded for a period of 30 minutes or until the end of the resuscitation, whichever occurred first. The recording investigator and the staff performing the resuscitation were blinded to all ETCO2 data.

Retrieval.  Differences between manual and computerized data collection are elaborated upon in Figure 1. The capnograph device used in the current study samples for CO2 every 50 msec. The sampled data are smoothed and corrected for potential and known interferences as part of the signal processing. The displayed ETCO2 is renewed per breath with the peak defined as the highest value measured after correction per breath.

Data were downloaded as ASCII files following disconnection of the patient from the monitor. The software used to download the sampled data was an early version of the current standard monitor data download software, but it was initially written for the purpose of this study. This software is designed to select and document the highest peak displayed within a recording time predefined by the user. A recording time of 30 seconds was determined for the purpose of this study. For example, if there were five breaths during the 30 recorded seconds, the documented ETCO2 would be that of the breath with the highest peak during this time period. The 30-second time frame was selected because it constitutes the default for standard documentation in capnograph devices. The concept was that no modifications should be required to repeat this study elsewhere with existing devices.

Following download, the data were transferred to MATLAB (The MathWorks, Inc., Natick, MA) and analyzed to derive trendline characteristics by a mathematician blinded to patient outcomes. The variables derived included peak ETCO2 values at 4, 5, 8, and 10 minutes; accumulating areas under the receiver operating characteristic (ROC) curve (AUCs) at 4, 10, and 20 minutes; trendline slope (rate of change per second) from 0 to 4, 5, 8, and 10 minutes; and the cumulative maxETCO2 (an average of all maxETCO2 measurements from time zero to the current point in time) up to 30 minutes.

The selection of these specific time points and the ETCO2 characteristics to be studied in this pilot were based upon clinical logic and previous publications that studied the relationship between real-time ETCO2 and survival from resuscitation. The clinical logic was as follows. In-hospital arrests are often witnessed, particularly during office hours.4 In a witnessed arrest, a time frame of 4–5 minutes with no flow is accepted as crucial for survival and neurologically intact survival. With the performance of optimal chest compressions, the cardiac output achieved is approximately 40% of normal cardiac output. Thus, assuming optimal chest compressions, a low-flow time of 8–10 minutes may again be significant for survival. Out-of-hospital termination of resuscitation is supported by the National Association of EMS Physicians for adult, nontraumatic cardiac arrest patients. However, the recommendation states that a full resuscitative effort be provided for at least 20–30 minutes prior to declaring the patient dead.3

Previous publications that studied the relationship between real-time ETCO2 and survival from resuscitation described either single5,6 or a few measurements (e.g., one a minute).7–10 Specific time points that were studied included the “initial” ETCO2,11–16 the ETCO2 at 4 minutes,17 the ETCO2 at 20 minutes,12,18 and the ETCO2 just before ROSC,5,6 or “final” ETCO2.14,15 Specific ETCO2 values that were sought included “mean” or “average” values7,8 and minimal13 and maximal9,13 values.

Study Endpoints.  Initial resuscitation was considered successful in case of ROSC. Further follow-up took place to ascertain whether the patient died in hospital or survived to hospital discharge.

The outcome measure was the presence of any relationship between the variables derived from ETCO2 trendlines and ROSC (primary) or survival to hospital discharge (SHD) (secondary).

Data Analysis

Descriptive statistics were used for patient demographics and event characteristics. Categorical variables (e.g., patients’ sex, witnessed status, and death) are expressed with their categories and the associated percentages. Continuous variables (e.g., patient age, time to arrival) are presented with their medians, interquartile range (IQR), and maximal and minimal values. Descriptive data were analyzed using SPSS 10.0 (SPSS, Inc., Chicago, IL).

Further analyses were performed using MATLAB (The MathWorks Inc., Natick, MA) as follows. The slopes and intercepts were calculated using linear curve fit. The linear fit captures the main trend of the data and enables grouping of cases for the purpose of comparison. Although it was clear that the small sample size would not allow drawing meaningful inferences regarding survival, data pooling was crucial for the purpose of understanding whether the characteristics of the ETCO2 data may inform the design of the ETCO2 analysis to generate a prediction model.

The possible differences in the independent variables (ETCO2 trendline characteristics) between the two dependent variable groups (ROSC/no ROSC and SHD/no SHD) were examined. Their significance (p-value) is presented. Significance testing was performed for proof-of-concept purposes only. Significance testing of the ETCO2 values and trendline characteristics of the two patient groups was performed without adjustments made for multiple comparisons.

Because all the study predictors were numerical, their distribution was not normal, and the study population was small, the Mann-Whitney test was selected to examine subgroup differences in capnography data. These data are presented in boxplots due to the distribution and the large amount of data to be shown.

Finally, the sensitivity and specificity of each variable or combination of variables for predicting ROSC/SHD were analyzed. Based on the data collected for this study, a limited model ruling out possible misdiagnosis of a patient with survival potential was created. Receiver operating characteristic (ROC) curves were used to examine the performance characteristics of each variable/combination of variables for predicting ROSC/SHD. ROC curve is a plot of 1- specificity versus sensitivity for all cutoff values in the range of the variable’s observed values. For each method, we estimated the AUC. AUC is used as a global index of test performance with an AUC = 0.5 indicating no discrimination ability and an AUC = 1 indicating perfect discrimination ability. An AUC = 0.8 indicates good discrimination ability.19 ROC curve confidence intervals (CIs) and p-values were calculated using a nonparametric model.

Results

The study population consisted of 30 patients with a median age of 74 years (IQR = 60–80 years; range = 16–92 years). Half were males, and most had been admitted for treatment of cardiac diseases. Most events (63%) were witnessed by the medical staff. The median time to arrival of the code team was 150 seconds (IQR = 105–255 seconds; range = 60–300 seconds). The rhythms most commonly documented at the time of connection to capnography were asystole and VF/VT. Additional patient demographics and event data are tabulated (Table 1). ROSC was achieved in 17 of 30 (57%) of the cases, and hospital discharge in six of 30 (20%).

Table 1. 
Study Population Demographics and Event Characteristics
Patient Characteristicsn%
  1. ICD-9 = International Classification of Diseases, 9th revision; PEA = pulseless electrical activity; VF/VT = ventricular fibrillation/ventricular tachycardia.

Male sex1550
ICD-9 admission diagnosis
 Diseases of the circulatory system1550
 Diseases of blood and blood-forming organs26.7
 Neoplasms13.3
 Diseases of the respiratory system26.7
 Septicemia620
 Mental disorders13.3
 Injury and poisoning26.7
 Diseases of the digestive system13.3
Event characteristics
Event location
 Surgical wards620
 Cardiology suite (including lab)310
 Internal medicine1240
 Intensive care units723.3
 Other26.7
Presenting rhythm
 Asystole930
 PEA620
 VT/VF723.3
 Bradyarrhythmia826.7
Rhythm at time of connection
 Asystole1240
 PEA620
 VF/VT1240
Witnessed–Yes1963.3

Capnography Data

ROSC.  Peak ETCO2 values in patients with and without ROSC differed significantly at 8 and 10 minutes after intubation, but not after 4 or 5 minutes (Figure 2). Patients without ROSC had significantly smaller (4 and 10 minutes) areas under the ETCO2 curve early than those with ROSC. However, this difference later disappeared, and considerable overlap existed (Figure 3).

Figure 2.

 The relationship between peak ETCO2 and ROSC. Box-and-whiskers plot for patients with (n = 17) and without (n = 13) ROSC for peak ETCO2 values at 4, 5, 8, and 10 minutes. Peak ETCO2 values in patients with and without ROSC only differed significantly at 8 (p = 0.042) and 10 minutes (p = 0.035) after intubation (marked with matching asterisks). ETCO2 = end tidal carbon dioxide; ROSC = return of spontaneous circulation.

Figure 3.

 The relationship between the area under the ETCO2 curve and ROSC. Box-and-whiskers plot for patients with (n = 17) and without (n = 13) ROSC for the AUC of the ETCO2 trendline at 4, 10, 20, and 30 minutes (presented on a logarithmic scale). Patients without ROSC had significantly smaller areas under the ETCO2 curve at 4 and 10 minutes than those with ROSC (p = 0.016 for both times, marked with matching asterisks) but this difference later disappeared. AUC = area under the curve; ETCO2 = end tidal carbon dioxide; ROSC = return of spontaneous circulation.

The slope of the ETCO2 trendline (Figure 4) and the cumulative maxETCO2 (Figure 5) demonstrated the most consistently significant differences between patients with and without ROSC with the least overlap. The ETCO2 trendline of patients without ROSC remained flat or sloped down, while that of patients with ROSC sloped up. The cumulative maxETCO2 > 20 at all time points between 5–10 minutes was the parameter most likely to predict ROSC (sensitivity = 0.88; specificity = 0.77; p < 0.001) (Figure 6), but even using this model, two patients who were predicted to not achieve ROSC did so.

Figure 4.

 The relationship between the slope of the ETCO2 trendline and ROSC. Box-and-whiskers plot for patients with (n = 17) and without (n = 13) ROSC for the slope (rate of change per second) of the ETCO2 trendline between 0 to 4, 5, 8, and 10 minutes. The slope of the ETCO2 trendline demonstrated consistently significant differences between patients with and without ROSC at 5 (p = 0.018), 8 (p = 0.016), and 10 (p = 0.016) minutes (marked with matching asterisks). ETCO2 = end tidal carbon dioxide; ROSC = return of spontaneous circulation.

Figure 5.

 The relationship between the cumulative maximal ETCO2 (maxETCO2) and ROSC. Box-and-whiskers plot for patients with (n = 17) and without (n = 13) ROSC for the cumulative maximal ETCO2 (average of all maximal ETCO2 measurements from time zero to the current point in time). The maxETCO2 was significantly different between patients with and without ROSC at all time points: 3.5 minutes, p = 0.001; 4 minutes, p = 0.0006; 4.5 minutes, p = 0.0006; 5 minutes, p = 0.0008; 5.5 minutes, p = 0.001; 6 minutes, p = 0.0012; 6.5 minutes, p = 0.0014; 7 minutes, p = 0.0014; 7.5 minutes, p = 0.0015; 8 minutes, p = 0.0011; 8.5 minutes, p = 0.0016; 9 minutes, p = 0.0011; 9.5 minutes, p = 0.006; 10 minutes, p = 0.0018; 10.5 minutes, p = 0.0024; 11 minutes, p = 0.0023. ETCO2 = end tidal carbon dioxide; ROSC = return of spontaneous circulation.

Figure 6.

 ROC curves for the variables best predicting ROSC. The ROC curve of variables best predicting return of spontaneous circulation included the ETCO2 slope between 0 and 8 minutes (triangles, AUC = 0.84, 95% CI = 0.69 to 0.98, p < 0.001), the cumulative maxETCO2 all time points measured between 5 and 10 minutes (circles, AUC = 0.84, 95% CI = 0.7 to 0.98, p < 0.001), and the combination of the two with a slope cutoff point at slope = 0 (diamonds, AUC = 0.95, 95% CI = 0.7 to 0.98, p < 0.001). The best cutoff point for the first parameter (slope) had a sensitivity 0.83, specificity 0.61, PPV 0.68, and NPV 0.8. The best cutoff point for the second parameter (maxETCO2) had a sensitivity of 0.88, specificity, 0.77, PPV 0.92, and NPV 0.69. For the combination we chose to use sensitivity = 1; at that point the specificity was 0.5, PPV 0.67 and NPV 1. AUC = area under the ROC curve; ETCO2 = end tidal carbon dioxide; NPV = negative predictive value; PPV = positive predictive value; ROC = receiver operator characteristic; ROSC = return of spontaneous circulation.

A system providing early notification of outcome should be characterized, as a necessary condition, with the possible highest sensitivity. To achieve a sensitivity of 1, a bivariate model was created based on a fixed condition of a false-negative rate of 0, allaying concerns regarding possible misdiagnosis of a patient with potential for recovery. The optimal resulting rule states that a patient will have ROSC if the cumulative maxETCO2 (at 5 to 10 minutes) is greater than 25 mm Hg or the slope between 0 to 8 minutes rises (Figure 6). The “cost” of the fixed condition to the model was a reduction in specificity to 0.5 compared to 0.77 prior to insertion of the fixed condition. In our cohort, this model correctly predicted the outcome of resuscitation in 70% of the cases within less than 10 minutes of intubation.

Survival to Hospital Discharge.  None of the examined variables predicted hospital discharge following in-hospital cardiac arrest and resuscitation.

Discussion

End tidal CO2 reflects tissue CO2 production, elimination, and flow through the lungs. It may therefore indirectly represent cardiac output. Our study demonstrates how capnography should be studied in the context of resuscitation. Rather than attempting to predict the outcome of resuscitation within minutes of intubation based on sporadic manual ETCO2 measurements, mathematical modeling of capnography should be further explored. When sufficiently understood, this tool could be integrated in multivariate indices that include variables known to be significant unto themselves (e.g., presenting rhythm and time to arrival). Capnography is becoming a standard of care after intubation; ETCO2 measurements are thus becoming readily available. These potentially valuable data should not be discarded.

The current study is the first to demonstrate how computerized sidestream ETCO2 data can be used for mathematical modeling of outcome prediction. Its results concur with most previous studies suggesting that capnography can facilitate prediction of ROSC. The added value of this study lies in demonstrating how data may inform the design of the ETCO2 analysis to generate a mathematical prediction model of outcome from resuscitation.

Similarly to our study, most previous studies were not powered to correlate capnography with SHD, resulting in either no data7,9,17 or conflicting results; both no correlation11 and a clear correlation5 have been described. Grmec and Kupnik,14 who studied 246 patients with out-of-hospital cardiac arrest, noted that all patients with SHD had initial ETCO2 values of higher than 10 mm Hg. Using this cutoff point, the sensitivity of initial ETCO2 values for predicting SHD was 1.0 for both initial (postintubation) and final ETCO2 measurements, with respective specificities of 0.8 and 0.92. Using a cutoff point of 15 mm Hg, sensitivity for predicting SHD remained 1.0 while specificity rose to 0.94 and 0.95 for initial and final measurements respectively. In a larger cohort, Grmec et al.15 demonstrated that both initial and final values of ETCO2 were significantly higher in patients with SHD compared with those who did not survive (initial 18.5 [SD ± 6.4] mm Hg vs. 10 [SD ± 6.4], final 27.8 [SD ± 8.6] mm Hg vs. 13.5 [SD ± 9.3], p < 0.0001 for both).

The quality of CPR is a critical determinant of survival in cardiac arrest. The fact that capnography is associated with lack of ROSC is not surprising if low ETCO2 values directly reflect ineffective CPR.20 Lacking tools to optimize CPR efforts, capnography could indicate that providers should perform better CPR, so that the chance of ROSC is realized. In the presence of alternative tools to direct the quality of CPR (e.g., mechanical chest compression or monitoring of impedance), capnography could become a patient-centered variable indicating the likelihood of success.

Clinical prediction rules recommending termination of resuscitation efforts are increasingly being studied, despite contention.21 However, these protocols usually do not incorporate capnography as a decision tool.

Limitations

This study has several limitations: it was performed in one medical center, and the sample was small. Multicenter trials include a larger number of participants from different geographic locations, possibly from a broader range of population groups, and also enable between-center comparisons. All of these increase the generalizability of the study. However, multicenter trials are also more likely to inadvertently include cases deviating from the study inclusion and treatment protocol. This study investigated only in-hospital arrests. This population is, on the one hand, sicker than the population suffering from prehospital cardiac arrest and, on the other hand, is better monitored and has a higher likelihood of early treatment.

Data were not collected on patients excluded from the study. However, historical data from the same hospital suggests that the cohort included in the current study does represent the overall in-hospital population of cardiac arrests.4 The cutoff points associated with a poor possibility of survival in this study were defined post hoc, a method associated with high likelihoods of type I errors. However, this study was designed only to describe how to study ETCO2 data, not what to infer from it. Indeed, it was not powered to detect correlations. To increase the generalizability of this study to CPR in any setting, its findings require substantiation in an expanded (preferably multicenter) in-hospital investigation and in a large prehospital study that would include a population differing in baseline severity of illness, treatment, and likelihood of survival.

Arterial blood gases were not sampled for PaCO2-ETCO2 gradient. Ventilation was performed manually during the resuscitation; changes in minute ventilation may have affected the study results.22 Administration of drugs such as adrenalin23–25 or bicarbonate26 may affect ETCO2, yet no data regarding the use of either drug are provided in this study. This should refine future work.

Conclusions

This preliminary study suggests that computerized end tidal carbon dioxide carries potential as a tool for early, real-time decision-making during some cardiac arrest resuscitations. Systematic mathematical modeling of capnography measurements recorded during resuscitation should be further investigated. Such data would provide benefit beyond that observed through sporadic manual end tidal carbon dioxide measurements.

Ancillary