Arterial and end-tidal CO2 gradient as a mortality predictor in critical care patients: a prospective observational study

Abstract Background Mortality prediction in critically ill patients is commonly performed with scoring systems such as SOFA and APACHE II. The arterial-to-end-tidal carbon dioxide gradient [P(a–et)CO₂], which can be obtained from routine blood gas analysis and capnography, may represent a simpler alternative. We investigated the association between this gradient and 28-day mortality in mechanically ventilated ICU patients. Methods This prospective observational study included 50 adult patients who had been intubated and had initiated invasive mechanical ventilation within the preceding 24 hours. Simultaneous arterial blood gas analysis and mainstream capnography were used to calculate the P(a–et)CO₂ gradient. Demographic data, comorbidities, vasopressor use, SOFA and APACHE II scores, PaO₂/FiO₂ ratio, ICU length of stay, ventilator days, and 28-day mortality were recorded. Statistical analysis included ROC curves and multivariable logistic regression. To ensure model robustness and prevent overfitting, internal validation was performed using 1,000 bootstrap resamples. Results Non-survivors had a significantly higher median P(a–et)CO₂ gradient than survivors (14.0 [range: 2.0–44.0] vs. 4.0 [range: -3.0–18.0] mmHg, p < 0.001). ROC analysis demonstrated a bootstrapped AUC of 0.890 (95% CI 0.761–0.980), with 12 mmHg as the optimal cut-off (sensitivity 0.78, specificity 0.94). In the internally validated multivariable analysis, the gradient remained independently associated with 28-day mortality (adjusted OR 4.21 per 5 mmHg increase, 95% CI 1.72–10.33, bootstrap p = 0.002). No correlation was found between the gradient and ICU stay or ventilator days. Conclusion The P(a–et)CO₂ gradient is a simple and valuable physiological marker independently associated with 28-day mortality in mechanically ventilated ICU patients. Its routine availability makes it a practical bedside adjunct for early risk assessment, though these findings warrant external validation in larger, multicenter cohorts. Trial registration The study was retrospectively registered at ClinicalTrials.gov (NCT05341258).