Mean 24-hour end-tidal carbon dioxide following diagnosis predicts mortality in patients with sepsis

Sepsis management has significantly improved over the past decades, with intensivists playing a pivotal role in its identification and treatment.^[1,2] However, resource constraints in large tertiary hospitals in China limit patient admissions, leading to overcrowding in the emergency departments (EDs) with critically ill patients.^[3] This highlights the urgent need for enhanced risk stratification and optimized sepsis management in emergency settings.

Lactate is a readily accessible marker for assessing tissue perfusion and predicting mortality in sepsis patients.^[4] However, its measurement requires blood tests, resulting in a delay of approximately two to three hours from ED triage to the initial result.^[5] End-tidal carbon dioxide (EtCO₂) is a real-time, non-invasive indicator that reflects global tissue perfusion and is proposed to be associated with lactate levels in patients with sepsis.^[6] Compared with other non-invasive perfusion assessments such as capillary refilling time, EtCO₂ enables continuous monitoring, making it a more promising indicator for risk stratification of sepsis in the ED.

While several studies have explored the associations between EtCO₂ levels and patient outcomes, most have focused on pre-resuscitation measurements, overlooking the overall trend of EtCO₂ during interventions.^[7,8] This study aimed to evaluate the prognostic value of the mean EtCO₂ (mEtCO₂) level within the first 24 h of sepsis diagnosis.

This is a retrospective cohort study that utilized the Medical Information Mart for Intensive Care (MIMIC)-IV database (version 2.2).^[9] Data extraction was performed by an authorized author (JYW) with access certification (No. 43427464). As MIMIC-IV data are de-identified, ethical approval was not required.

The inclusion criteria included adult ICU patients (aged ≥ 18 years) with sepsis who had at least two measurements of EtCO₂ within 24 h after sepsis diagnosis. Sepsis and septic shock were diagnosed according to the Sepsis-3 definition.^[10] Patients were excluded if they remained in the ICU for less than 24 h following sepsis diagnosis. For patients with multiple ICU admissions, only the first admission was included.

Data extraction and processing were performed using Navicat Premium (version 16.1.3) and Stata (version 17.0). EtCO₂ measurements were recorded within 24 h of sepsis diagnosis. The initial EtCO₂ level (iEtCO₂) was defined as the first measurement. The mEtCO₂ was calculated as the average of the recorded values over 24 h. Baseline characteristics were documented, with further details provided in the Supplementary Files.

The primary outcome was 28-day mortality. The secondary outcomes included the number of ventilator-free days, vasopressor-free days, and ICU length of stay (LOS) within 28 d.

Statistical analyses were performed via SPSS 26.0 and GraphPad (version 10.2.2). Continuous variables were presented either as mean ± standard deviation or median (interquartile ranges), depending on the normality of the distribution assessed by the Shapiro-Wilk test. Categorical variables were presented as frequencies and percentages. Group comparisons were made using Student’s t-test or the Mann-Whitney U test for continuous variables, and the Chi-square test or Fisher’s exact test for categorical variables. To address missing data for baseline characteristics, median imputation was used for missing rates below 5%, whereas multivariate imputation by chained equations was applied for missing rates between 5% and 15% using R language.

The dataset was randomly divided into training and testing cohorts a 7 to 3 ratio. The training cohort was used to determine the optimal mEtCO₂ cut-off value for predicting 28-day mortality. Patients in the testing cohort were subsequently stratified into high and low mEtCO₂ categories based on this cut-off, and their clinical outcomes were compared. Logistic regression models were employed to determine the association between the level of mEtCO₂ and 28-day mortality by adjusting adjusted for potential confounding factors, including age, gender, Sequential Organ Failure Assessment (SOFA) score, Simplified Acute Physiology Score (SAPS II), comorbidities, interventions, mean arterial pressure (MAP), and mean heart rate (HR) over 24 h. Subgroups were stratified by age (≤ 65 years/ > 65 years), gender (male/ female), SOFA score (≤ 5/ > 5), SAPS II (≤ 45 / > 45), and the presence or absence of shock at baseline. Subgroup analyses were conducted to evaluate the predictive efficacy of mEtCO₂ in these different subgroups. In addition, a sensitivity analysis was performed to examine the impact of lower EtCO₂ (< 30 mmHg) frequencies on mortality. The details of logistic regression, subgroup, and sensitivity analyses were provided in the Supplementary Files. The receiver-operating characteristic (ROC) curve was used to assess the ability of mEtCO₂ to predict 28-day mortality. A two-sided P-value < 0.05 was considered statistically significant.

After reviewed 31,983 records of ICU admission for sepsis, a total of 2,482 patients were included (Supplementary Figure 1). The dataset was randomly divided into a training cohort of 1,732 patients and a testing cohort of 750 patients, with negligible differences in baseline characteristics between the two cohorts (Supplementary Table 1).

In the training cohort, 453 patients died at 28-day follow-up. Non-survivors had significantly lower mEtCO₂ and iEtCO₂ levels than survivors did (Table 1). ROC analyses revealed that mEtCO₂ had a superior predictive performance for mortality than did iEtCO₂, with areas under the curve (AUC) of 0.687 (95% CI 0.658-0.715) and 0.631 (95% CI 0.600-0.661), respectively (P=0.006) (Supplementary Figure 2). The optimal mEtCO₂ cut-off was 30 mmHg, with a sensitivity of 74.2% and a specificity of 52.3%.

Table 1.

Baseline characteristics of patients in the training cohort

Variables	Survival (n=1,279)	Non-survival (n=453)	P-value
Demographics
Age, years, mean±SD	62 ± 16	68 ± 15	<0.001
Male gender, n (%)	759 (59.3)	246 (54.3)	0.062
Comorbidities, n (%)
Cardiac disease	437 (34.2)	171 (37.7)	0.170
Chronic pulmonary disease	326 (25.5)	118 (26.0)	0.815
Liver disease	185 (14.5)	85 (18.8)	0.030
Cerebrovascular disease	166 (13.0)	90 (19.9)	<0.001
CKD	255 (19.9)	100 (22.1)	0.333
Severity of illness, mean±SD
SOFA score	4 ± 2	5 ± 3	<0.001
SAPS II	43 ± 14	55 ±15	<0.001
EtCO2 measurement
mEtCO2, mmHg, mean±SD	36 ± 6	31 ± 7	<0.001
iEtCO2, mmHg, mean±SD	35 ± 7	32 ± 7	0.001
Measurement frequency, median (IQR)	4 (2-5)	4 (3-5)	0.863
Vital signs
MAP, mmHg, median (IQR)	60 (53-69)	57 (49-68)	0.002
HR, beats per minute, mean±SD	103 ± 31	110 ± 32	<0.001
Temperature, °C, mean±SD	36.6 ± 1.3	36.1 ± 1.5	<0.001
RR, breaths per minute, mean±SD	24 ± 12	25 ± 12	0.041
Laboratory results, mean±SD
HCO3-, mmol/L	20.1 ± 6.0	17.2 ± 6.1	<0.001
Lactate, mmol/L	2.3 ± 1.6	4.0 ± 3.4	<0.001
Treatments, n (%)
Vasopressor	522 (40.8)	215 (47.5)	0.014
CRRT	35 (2.7)	43 (9.5)	<0.001
IMV	1,250 (97.7)	445 (98.2)	0.526

SD: standard deviation; IQR: interquartile range; CKD: chronic kidney disease; SOFA: Sequential Organ Failure Assessment; SAPS II: Simplified Acute Physiology Score II; MAP: mean arterial pressure; HR: heart rate; RR: respiratory rate; mEtCO₂: mean end-tidal carbon dioxide; iEtCO₂: initial end-tidal carbon dioxide; CRRT: continuous renal replacement therapy; IMV: invasive mechanical ventilation.

In the testing cohort, 173 (23%) patients were classified as having low mEtCO₂ and 577 (77%) were classified as having high mEtCO₂, based on the 30 mmHg threshold. Patients in the low mEtCO₂ group were older, had higher SPAS II scores, and had worse vital signs and laboratory results (Supplementary Table 2). A total of 82 of the 173 patients (47.4%) in the low mEtCO₂ group and 118 of the 577 patients (20.5%) in the high mEtCO₂ group died at 28-day of follow-up. Compared with patients with high mEtCO₂, those with low mEtCO₂ had significantly fewer ventilator-free days (3 [0-23] d vs. 22 [8–26] d, P<0.001) and vasopressor-free days (21 [0-28] d vs. 28 [20–28] d, P<0.001). The ICU LOS did not differ between the two groups (5 [3–10] d vs. 5 [3–11] d, P=0.088) (Table 2). After adjusting for multiple covariates, mEtCO₂ < 30 mmHg remained significantly associated with increased 28-day mortality (OR 2.845; 95%CI 1.920-4.215; P<0.001) (Supplementary Table 3). mEtCO₂ also showed a stronger association with mortality than other hemodynamic parameters such as the MAP and HR (Supplementary Table 4). Subgroup analysis corroborated the main findings, with all subgroups demonstrating a significant association between low mEtCO₂ and 28-day mortality (Supplementary Figure 3). Moreover, the sensitivity analysis indicated that prolonged exposure to low mEtCO₂ levels was associated with an increased risk of death (Supplementary Table 5).

Table 2.

Clinical outcomes of patients in the testing cohort

Outcomes	Low mEtCO2 (n=173)	High mEtCO2 (n=577)	P-value
28-day mortality, n (%)	82 (47.4)	118 (20.5)	<0.001
Ventilator-free days, d, median (IQR)	3 (0-23)	22 (8-26)	<0.001
Vasopressor-free days, d, median (IQR)	21 (0-28)	28 (20-28)	<0.001
ICU LOS, d, median (IQR)	5 (3-10)	5 (3-11)	0.088

ICU: intensive care unit; IQR: interquartile range; LOS: length of stay; mEtCO₂: mean end-tidal carbon dioxide. Low mEtCO₂ was defined as the mEtCO₂< 30 mmHg within 24 h of sepsis diagnosis; High mEtCO₂ was defined as the mEtCO₂ ≥ 30 mmHg within 24 h of sepsis diagnosis.

ROC analysis demonstrated that mEtCO₂ had comparable predictive performance to mean lactate, with an AUC of 0.688 (95% CI 0.644-0.731). The optimal mEtCO₂ cut-off in the testing cohort was 29 mmHg, which was closely aligned with the 30 mmHg threshold in the training cohort. The SAPS II score was also a significant predictor of 28-day mortality, with an AUC of 0.713 (95% CI 0.672-0.755) (Figure 1). The predictive performance of the combination of mEtCO₂ and SAPS II was significantly superior to that of either variable alone, yielding an AUC of 0.760 (95% CI 0.723-0.800) (P<0.001 for mEtCO₂, P =0.005 for SAPS II).

Figure 1. ROC curves for predicting 28-day mortality in the testing cohort. mEtCO₂: mean end-tidal carbon dioxide; SAPS II: Simplified Acute Physiology Score II; AUC: areas under the curve. The predicted probability of mortality derived from the combination of mEtCO₂ and SAPS II used a binary logistic regression model.

Our study demonstrated that in patients with sepsis, an mEtCO₂ < 30 mmHg within 24 h of diagnosis was significantly associated with increased 28-day mortality. In addition, patients with lower mEtCO₂ levels require prolonged support from ventilators and vasopressors.

From a physiological perspective, CO₂ is produced through aerobic metabolism and is primarily transported as bicarbonate to the lungs for exhalation. Thus, EtCO₂ may serve as a useful indicator of cardiac output (CO). Previous studies have shown a significant correlation between EtCO₂ levels and CO,^[11,12] indicating that EtCO₂ may provide insights into hemodynamic status, particularly in settings lacking of advanced monitoring equipment.

While prior research has emphasized the prognostic value of EtCO₂ in sepsis,^[8] our study enhanced these findings by incorporating temporal trends in EtCO₂ measurement. The superior predictive performance of mEtCO₂ over iEtCO₂ further highlight the necessity of dynamic monitoring. Nevertheless, the prognostic utility of EtCO₂ alone remains limited. As a continuously measurable and non-invasive parameter, however, EtCO₂ still holds significant potential as a promising indicator for sepsis assessment in the ED.

SAPS II is appropriate for EDs because of its simplicity, incorporating critical factors such as age, comorbidities and vital signs.^[13] Our study suggested that combining the SAPS II with EtCO₂ provides a more comprehensive prognostic model than any of the variables alone.

The study has several limitations. First, although we aim to assess the performance of EtCO₂ in emergency settings, our study relied on ICU-based electronic medical data. Despite this, the focus on sepsis diagnosis within 24 h made our findings applicable to institutions facing ED congestion. Second, fewer than 2% of patients with available EtCO₂ values did not receive invasive mechanical ventilation (IMV), limiting our analysis to non-ventilated patients. Further studies are needed to clarify the efficacy of EtCO₂ in these patients. Third, owing to the intermittent and irregular measurement of EtCO₂, we were unable to assess the potential advantages of continuous monitoring. Prospective studies are needed to investigate the role of dynamic, continuous EtCO₂ measurements in sepsis management.

In conclusion, a 24-hour mEtCO₂ < 30 mmHg was significantly associated with 28-day mortality in patients with sepsis. The real-time and non-invasive nature of EtCO₂ measurement makes it a potential indicator for sepsis risk stratification in the ED.

Funding: This work was supported by grants from the Chinese Academy of Medical Sciences Innovation Fund for Medical Sciences (2021-I2M-1-062, to BD).

Ethical approval: Ethical consent was not required in this study since the MIMIC-IV data were deidentified.

Conflicts of interest: The authors have no relevant financial or non-financial interests to disclose.

Author contributions: All authors contributed to the study conception and design. Material preparation, data collection and analysis were performed by JYW and JX. The first draft of the manuscript was written by JYW. LW and BD provided critical revisions to the manuscript. All authors read and approved the final manuscript.

All the supplementary files in this paper are available at http://wjem.com.cn.