Diagnostic Value of End-Tidal Carbon Dioxide Combined With Lactate for Early Detection of Sepsis in Prehospital Patients: A Prospective Cohort Study

Introduction

Sepsis is a systemic inflammatory response syndrome triggered by infection, characterized by rapid progression and high mortality [1,2]. Early detection and intervention in the prehospital setting – the first point of care for septic patients – are crucial for improving outcomes. Currently, prehospital sepsis assessment relies predominantly on clinical scoring tools such as SIRS, qSOFA, and NEWS2, which are based on vital signs and clinical judgment. While useful, these tools lack real-time assessment of perfusion status and metabolic derangement, resulting in limited sensitivity and specificity for early sepsis detection [3].

In recent years, biomarkers have been explored to address this gap. Lactate measurement has been proposed for early risk stratification, with evidence showing that prehospital lactate testing can enhance the identification of suspected sepsis patients at elevated mortality risk during emergency transport [4]. However, hyperlactatemia occurs in only about 65% of septic shock cases, indicating that lactate alone may not consistently reflect tissue hypoxia across all patients [5]. This limitation underscores the need for complementary markers that more directly capture perfusion and metabolic dysfunction.

End-tidal carbon dioxide (ETCO2) monitoring has emerged as a promising tool in emergency medicine, offering non-invasive, continuous insight into ventilation–perfusion coupling [6,7]. Studies suggest its utility in early sepsis recognition and risk assessment, with ETCO2 levels correlating with lactate and SOFA scores [8–10]. Notably, an ETCO2 ≤25 mmHg has shown superior predictive value for mortality compared to qSOFA in the prehospital setting. Furthermore, meta-analyses suggest that combining multiple biomarkers – particularly those from different sepsis-related pathways – can enhance diagnostic accuracy beyond that of any single marker [11,12].

Nevertheless, prior studies have not fully resolved key knowledge gaps, including prehospital workflow feasibility, threshold performance in EMS populations, and the incremental value of adding lactate to ETCO2 alone. This study aimed to evaluate the combined use of ETCO2 and lactate for early identification and risk stratification of sepsis in the prehospital setting. This dual-marker strategy could play a role in integrating ventilation–perfusion coupling and tissue hypoxia into early sepsis recognition prior to hospital arrival.

Material and Methods

ETHICS STATEMENT:

This study strictly adhered to the principles of the Declaration of Helsinki. The study protocol was reviewed and approved by the Ethics Committee of the First People’s Hospital of Changde, Hunan Province, China (approval number: YX-2023-417-01; approval date: January 2023).

During the informed consent process, ethics procedures were followed. For patients with full decision-making capacity, written informed consent was obtained directly from them. In emergency situations such as prehospital care, where patients were critically ill and unable to express their wishes, informed consent was obtained from the patient’s immediate family members or legal representatives in accordance with relevant ethics guidelines and regulations.

STUDY DESIGN:

A single-center prospective observational cohort study was conducted, consecutively enrolling patients with suspected sepsis from the prehospital settings of Changde, China between January 2023 and January 2025. Data were collected continuously and non-selectively for these patients at Changde Hospital. The primary objective of the study was to evaluate the performance of ETCO₂ combined with lactate for diagnosing sepsis in the prehospital setting. It should be noted that prehospital biomarker studies are vulnerable to selection bias and confounding.

STUDY POPULATION:

During the study period, 353 patients with suspected sepsis were screened in the prehospital setting. The participant selection criteria were: Adult patients (≥18 years) with suspected infection who met ≥2 SIRS criteria (temperature >38°C or <36°C, heart rate >90 beats/min, or respiratory rate >20 breaths/min) were enrolled. A prehospital sepsis alert was triggered only when prehospital personnel confirmed the infection based on clinical assessment and the patient simultaneously met the SIRS criteria.

The selection of SIRS criteria for prehospital screening is based on the following key rationales.

The sensitivity, specificity, and practicality of qSOFA in the early identification of sepsis remain controversial. The 2021 Surviving Sepsis Campaign (SSC) guidelines recommend prioritizing SIRS over qSOFA for screening purposes [13]. qSOFA demonstrates low sensitivity in predicting sepsis-related mortality, whereas SIRS achieves significantly higher sensitivity. In prehospital settings, high sensitivity is critical to avoid missed diagnoses and the resulting rapid clinical deterioration [14].

Exclusion criteria included patients who were under 18 years of age, as well as those with cardiac arrest, do-not-resuscitate (DNR) orders, terminal illnesses, significant trauma, or missing data for ETCO2 and/or lactate. Ultimately, 327 were enrolled in the study. We followed up with all patients. Of these, 279 met the diagnostic criteria for either sepsis or septic shock according to the Sepsis-3.0 guidelines (Figure 1).

The total sample size was calculated using the formula based on sensitivity [15]. The preliminary trial demonstrated that ETCO2 predicted sepsis with 90.3% sensitivity and 80.5% specificity, which aligns with findings previously reported in the literature [6,7]. Hence, with an absolute error of 10% and a type I error of 0.05, the minimum total sample size was 62. The sample size was reported and justified based on power calculations, ensuring the results are statistically significant.

DATA SOURCES AND DATA INTEGRITY:

The data in this study primarily comprise the following 3 categories, which were rigorously linked and verified to ensure data integrity:

To ensure the accuracy and traceability of data linkage, all data were matched using patient unique ID and timestamp as key variables. The data extraction and verification process was independently performed by 2 researchers.

EMS TRAINING:

All prehospital emergency medical personnel involved in this study received uniform training on the research protocol. The training content included:

MEASUREMENT:

ETCO2 was measured and recorded by respiratory therapy staff with a sidestream sampling nasal cannula (Phillips sidestream ETCO2 monitor). The ETCO2 value was recorded after the CO2 waveform peak stabilized over 3 to 5 consecutive breaths. For patients requiring emergency intubation due to respiratory distress, the ETCO2 value was documented immediately prior to mechanical ventilation, once the CO2 peak remained consistent across 3 to 5 breaths. Arterial blood gas samples were collected spontaneously and analyzed using a POCT devices to measure lactate. Hyperlactatemia defined as ≥2 mmol/L [16]. Initial prehospital vital signs recorded by emergency medicine personal included temperature, heart rate (HR), systolic blood pressure (SBP), respiratory rate (RR), ETCO2, and lactate. Baseline characteristics such as sex, age, and comorbidities were documented in the prehospital care records. Transport time was defined as the duration from the scene of the emergency call to the arrival at the destination hospital. Initial ED laboratory data, including complete blood count, urine culture, blood culture, and respiratory viral panel, were recorded.

OUTCOMES:

The primary outcome variable was diagnosed sepsis and septic shock. The secondary outcome was ICU admissions and in-hospital mortality. Sepsis and septic shock were defined based on the Third International Consensus Definitions for Sepsis and Septic Shock (Sepsis-3) [16]. Sepsis 3.0 proposes that sepsis can be diagnosed when a patient meets the criteria of “infection+SOFA score ≥2 points”. Septic shock is diagnosed when, after adequate fluid resuscitation, the patient requires vasopressors to maintain a mean arterial pressure ≥65 mmHg and has a serum lactate level 2 mmol/L.

The final diagnosis of sepsis or septic shock for study purposes was not based on a single clinician’s assessment but was determined through a formal adjudication process. This process involved a joint review of the patient’s case by both ED physicians and ICU physicians. Their consensus, following the Sepsis-3 definitions, constituted the final diagnosis used for the primary outcome.

STATISTICAL ANALYSIS:

The statistical analysis was performed using SPSS 22.0 software. Categorical variables are expressed as proportions with frequencies and percentages, and continuous variables are expressed as mean ±SD and medians with interquartile ranges (IQRs).

The Wilcoxon rank sum test was used to compare ETCO₂ and lactate between sepsis and septic shock. ETCO₂ was analyzed as a continuous variable (per 5-mmHg decrease), and both linear and nonlinear relationships with clinical outcomes were assessed. Lactate was analyzed as a continuous variable (per 0.5 mmol/L increase), and both linear and nonlinear relationships with clinical outcomes were assessed.

We constructed a multivariable logistic regression model to examine the independent association between ETCO₂, lactate, and clinical outcomes. The primary independent variables were ETCO₂ and arterial lactate levels measured in prehospital settings, both of which were included in the model as continuous variables. The multivariate regression model was adjusted for demographic factors (ie, age and sex), comorbidities, infection source, transport time, and first recorded vital parameters at first contact with an emergency medicine physician (ie, SBP, HR, and oxygen saturation as measured by pulse oximetry). We excluded patients with missing data from the multivariate model. We assessed the model’s fitting performance using a residual plot.

We used 2 ETCO2 thresholds – ETCO2 ≤25 mmHg and ETCO2 ≤30 mmHg – to identify prehospital patients at high risk of sepsis. This approach is consistent with previous reports [6,8]. Sensitivities and specificities with 95% confidence intervals were calculated for each possible integer cut-point on the ETCO2 and the results were plotted in a receiver operator characteristic (ROC) curve. The cut-points were predefined and internally evaluated only. The area under the curve (AUC) with 95% confidence intervals was calculated to evaluate the diagnostic performance of ETCO2 for sepsis. The areas under the ROC curves were compared using the method described by DeLong.

The “combined model” is a prespecified binary decision rule (ETCO₂ ≤25 mmHg and lactate ≥2 mmol/L).The thresholds were prespecified rather than optimized post hoc. First, independent cut-off values (ie, thresholds for determining abnormality) were established for ETCO₂ and lactate, with ETCO₂ ≤25 mmHg and lactate ≥2.0 mmol/L. A case was classified as positive only when both ETCO₂ ≤25 mmHg and lactate ≥2.0 mmol/L were met. Our study employed a simple binary combination rule with predefined thresholds and calculated the sensitivity and specificity of this established rule to evaluate its predictive value. We evaluated the diagnostic value of combining ETCO₂ with hyperlactatemia for prehospital sepsis detection by integrating ETCO₂ cut-off values at varying thresholds with hyperlactatemia and calculating sensitivity, specificity, and 95% CIs for each diagnostic approach.

Results

PATIENT CHARACTERISTICS:

The mean ETCO2 was 23±4 mmHg in patients with sepsis and 14±3 mmHg in those with septic shock (P=0.004). The mean lactate was 2.5±1.6 mmol/L in patients with sepsis and 3.9±2.1mmol/L in patients with septic shock (P<0.001). Compared to patients with sepsis, those with septic shock had significantly higher in-hospital mortality (31% vs 14%) (P=0.010) and ICU admission rates (47% vs 22%) (P=0.030). However, there was no significant difference between sepsis and septic shock in terms of age, sex, comorbidities, white blood cell count, platelet count, or source of infection (Table 1). This discrepancy can be attributed to 2 factors: First, all patients were pre-selected based on SIRS criteria, which may have homogenized the cohorts at baseline, and second, the study had limited statistical power to detect subgroup differences.

:

Prehospital ETCO2 showed a linear relationship with both septic shock (OR: 1.07 [95% CI, 1.01–1.15] per 5 mmHg decrease) and in-hospital mortality (OR: 1.04 [95% CI, 1.01–1.12] per 5-mmHg decrease). Similarly, prehospital lactate was linearly associated with septic shock (OR: 1.10 [95% CI, 1.07–1.21] per 0.5-mmol/L increase) and in-hospital mortality (OR: 1.08 [95% CI, 1.01–1.15] per 0.5 mmol/L increase) (Table 2). However, no linear relationship was observed between prehospital ETCO2/lactate and ICU admission. The residual plots for each model are shown in Figure 2. A random pattern exhibited in the residual plot demonstrates the adequacy of the model specification.

:

Sensitivities and specificities (with 95% confidence intervals) for various cut-points of ETCO2 were calculated and presented using a receiver operating characteristic (ROC) curve. The AUC for ETCO2 ≤25 mmHg and ETCO2 ≤30 mmHg to predicting prehospital sepsis was 0.781 (95% CI 0.738–0.846) and 0.679 (95% CI 0.574–0.784), respectively. The area under ROC curve for ETCO2 ≤25 mmHg was significantly greater than ETCO2 ≤30 mmHg (Z=2.360, P=0.018) (Figure 3). This indicates that setting the threshold at ≤25 mmHg provides better overall discriminative ability for the model.

DIAGNOSTIC PERFORMANCE OF THE COMBINED ASSESSMENT:

Table 3 reveals that the dual-marker approach (ETCO2 ≤25 mmHg+hyperlactatemia) achieves a superior balance by significantly reducing both false negatives and false positives compared to single-marker strategies. Reduction in false negatives: The sensitivity of the combined rule (ETCO2 ≤25+hyperlactatemia) reached 89.9%, substantially higher than that of ETCO2 ≤25 mmHg alone (82.9%) and ETCO2 ≤30 mmHg alone (77.3%). Reduction in false positives: Crucially, this gain in sensitivity did not compromise specificity. The specificity of the combined assessment (91.7%) was markedly higher than that of ETCO2 ≤25 mmHg alone (81.7%) and ETCO2 ≤30 mmHg alone (75.1%).

The positive and negative predictive values were 98.4% (95% CI: 96.2–99.4%) and 61.1% (95% CI: 50.0–71.3%), respectively (Table 4). PPV and NPV are directly influenced by disease prevalence. The prevalence of Sepsis-3–positive cases in our cohort of 327 prehospital alerts was high (85.3%, 279/327). This high baseline prevalence is the key factor contextualizing the notably high PPV observed in this study.

Discussion

Sepsis remains a significant global health challenge due to its nonspecific early presentation, high rates of delayed diagnosis, and frequent misdiagnosis. With the sepsis management bundle now condensed to a 1-h window[17], the demand for rapid and reliable early screening tools has intensified. However, most existing assessment tools rely on complex laboratory parameters suited for in-hospital settings, lacking practicality in prehospital environments. Our study addresses this gap by exploring a feasible early screening strategy for prehospital sepsis recognition and proposing a rapid scoring system that can be integrated into clinical criteria for identifying sepsis and septic shock.

We found that combining ETCO2 ≤25 mmHg with lactate ≥2 mmol/L offers high sensitivity and specificity for early prehospital sepsis identification. These markers serve as critical “warning signals,” prompting prehospital personnel to suspect sepsis earlier and initiate timely interventions. It is important to frame the combined ETCO2–lactate assessment not as a definitive diagnostic test, but as a prehospital decision-support tool. Its primary value lies in augmenting clinical judgment by providing objective, physiological data to flag patients at high risk of deterioration. This tool can guide severity-based escalation during transport. For example, a positive result could trigger protocolized actions such as early fluids and pre-alerting the receiving hospital’s ICU or resuscitation team.

Throughout the progression of sepsis, lactate is the most widely recognized biomarker for tissue hypoperfusion and shock. However, triaging septic patients based on lactate remains controversial for the following reasons. (1) Low etiological specificity: Hyperlactatemia is not unique to sepsis but is a common feature of all types of shock (hypovolemic, cardiogenic, obstructive, distributive) [18,19]. (2) Reliance on a single indicator may lead to clinical misjudgment: A post hoc analysis of the ANDROMEDA-SHOCK study found that in patients with septic shock and normal capillary refill time, lactate-guided resuscitation was associated with more organ dysfunction, suggesting potential risks of overtreatment [20]. By integrating ETCO2 monitoring, which provides real-time insights into ventilation and perfusion, our dual-marker strategy enhances diagnostic accuracy and offers functional complementarity to lactate measurement.

Previous research has confirmed that ETCO2 ≤25 mmHg has greater utility than ETCO2 ≤30 mmHg in predicting both the accuracy and severity of prehospital suspected sepsis patients[10], which aligns with the findings of this study. Furthermore, we found that each 5 mmHg decrease in prehospital ETCO2 was associated with increased in-hospital mortality. A decreased ETCO2 reflects reduced efficiency of CO2 transport from tissues to the lungs. In sepsis patients, microcirculatory dysfunction and increased alveolar dead space ventilation are the primary pathophysiological mechanisms [21,22]. This study confirms and reinforces that, in the prehospital setting, ETCO2 should be considered an important parameter for the early recognition of sepsis. However, it is important to acknowledge that a low ETCO2 reading is not specific to sepsis. In the prehospital setting, conditions such as anxiety-induced hyperventilation, acute exacerbations of pulmonary diseases (eg, asthma, COPD), or even sampling artifacts can also result in reduced ETCO2. This inherent lack of specificity underscores why our dual-marker approach – requiring concurrent evidence of metabolic stress (lactate ≥2 mmol/L) – is crucial for improving the specificity of screening and the accuracy of decision-making.

Recent studies have shown that performing lactate analysis in prehospital settings for patients with suspected sepsis can significantly enhance early identification of those at high mortality risk, particularly among patients not flagged as high risk by existing risk stratification scores, such as RETTS and NEWS2 [23], which focused on assessing the risk of mortality in prehospital sepsis patients using lactate levels, whereas our study explored the early identification of suspected sepsis patients in the prehospital setting based on lactate combined with ETCO2. Olander et al proposed using lactate as a biomarker for early identification of sepsis in prehospital settings [24]. Their findings indicated that a lactate level of ≥2 mmol/L had limited predictive value for sepsis in the prehospital setting, suggesting that, in prehospital environments, lactate alone cannot effectively predict sepsis. However, its widespread elevation indicates potential utility in early clinical decision-making, warranting further investigation into its role in multimodal assessments. Unlike the study by Olander, our study used the combined method of ETCO2 and lactate for the prehospital early identification of sepsis, proposing a solution for the exploration of early sepsis screening tools suitable for prehospital settings.

Currently, the qSOFA score is considered to have high specificity but low sensitivity as a screening tool for non-ICU sepsis patients [25], which would result in many missed diagnoses and is not conducive to early prehospital sepsis screening. Our study prospectively evaluated the feasibility, accuracy, and impact on triage decision-making of using ETCO2 combined with lactate for early warning of prehospital sepsis identification. Our findings support refining existing triage systems such as qSOFA or NEWS2 [26]. Incorporating “lactate ≥2.0 mmol/L” and “ETCO2 ≤25 mmHg: as additional scoring criteria – resulting in an adapted score such as qSOFA-L/E – can improve risk stratification. For instance, patients with initially low traditional scores but abnormal dual-marker results would receive higher scores, prompting routing into higher-acuity care pathways. This facilitates destination decision-making, where high-risk patients can be routed directly to the ICU, while those with normal markers may be directed to general emergency departments. This integration exemplifies the combined method as a bridging tool that connects prehospital assessment to in-hospital response, ensuring a continuum of care aimed at early intervention. However, we explicitly note that protocol-level actions suggested by these findings – such as direct ICU transfer – remain hypotheses for future implementation or interventional trials, as we did not assess whether acting on these biomarkers translates to improved patient outcomes. Similar modifications, such as adding procalcitonin to create a PqSOFA score [27], have shown improved screening sensitivity, reinforcing the value of biomarker integration.

The study has 5 limitations: (1) Potential measurement errors (sidestream vs mainstream ETCO2). The use of sidestream ETCO2 monitoring, while practical for the prehospital setting, is susceptible to inherent inaccuracies. These inaccuracies can be exacerbated in non-sepsis conditions (eg, hyperventilation, pulmonary disease, anxiety, sampling limitations), potentially leading to falsely low readings and affecting the specificity of ETCO2-based screening. In patients with high respiratory rates or low tidal volumes – common in sepsis – the device may fail to capture a true alveolar plateau, potentially underestimating the ETCO2 value. Additionally, factors such as gas dilution in the sampling tube and moisture entrapment could introduce further measurement error. (2) Timing differences between ETCO2 and lactate sampling. The dynamic nature of prehospital care meant that the continuous ETCO2 monitoring and the single-point lactate measurement were not perfectly synchronized. This asynchrony is particularly relevant when considering transient non-sepsis causes of low ETCO2 (eg, brief hyperventilation), as it can potentially affect the combined method’s utility in real-time decision-making. In the dynamic prehospital environment, this combined method should serve as a decision-support tool to initiate a comprehensive clinical assessment, rather than an absolute diagnostic endpoint. (3) Possible device calibration variability. As data were collected from multiple emergency teams and units, potential variability exists. Although standardized protocols were followed, we cannot exclude subtle influences from inter-device calibration differences in both ETCO2 capnometers and lactate meters, as well as minor procedural variations among different operators during blood sampling and device setup. (4) Our study is based on a single-center design and lacks external validation. Multicenter research is required for further verification. (5) Our findings should not yet be used to inform protocol-level changes.

Conclusions

The prehospital application of combined ETCO2 and lactate measurement demonstrates promising discriminatory performance for early physiologic risk stratification within a single-center EMS cohort. Multicenter prospective validation with prespecified thresholds and implementation-focused endpoints is required before recommending integration into standardized EMS sepsis care.