Blood gas readings at the emergency door flag who is most likely to die

When critically ill patients are rushed into a hospital resuscitation room, one of the first tests they undergo is a point-of-care blood gas analysis, a bedside measurement that returns values for acidity, bicarbonate, lactate, glucose and hemoglobin within minutes. A retrospective cohort study from a German university hospital now reports that three of these rapidly available readings—anemia, acidosis and low bicarbonate—were independently associated with in-hospital mortality in critically ill non-trauma patients, even after adjustment for age, sex and out-of-hospital cardiac arrest. The study, published in BMC Emergency Medicine, analyzed 542 adults triaged as ESI level I or II who were treated in the resuscitation room of the Emergency Department of the University Hospital Regensburg between January 2016 and August 2017.

The stakes in this population are high. Patients allocated to the resuscitation room were identified by emergency medical services as having at least one immediately life-threatening condition, such as airway compromise, respiratory failure, hemodynamic instability or a critical neurological emergency. Most patients in the cohort presented with cardiac disease (48 percent), respiratory disease (15 percent) or intoxication (12 percent) as the underlying diagnosis. Of the 879 patients treated in a resuscitation room during the study period, 542, or 61.7 percent, met the inclusion criteria, and the in-hospital mortality rate in the analyzed cohort was 20 percent, corresponding to 110 deaths.

The researchers collected the first blood gas analysis obtained in the resuscitation room along with baseline clinical data from electronic medical records. Both venous and arterial samples were analyzed with equal weighting, because previous literature supports acceptable clinical concordance between the two sampling methods for the parameters assessed. Partial pressure of oxygen was excluded, since venous and arterial measurements are not interchangeable, and carbon dioxide partial pressure was not treated as a primary metabolic marker because its interpretation depends on ventilatory status, which was not consistently documented. Continuous variables were categorized using established clinical reference ranges, and glucose was dichotomized using mortality-associated breakpoints reported by Bernhard and colleagues in critically ill emergency patients.

In the multivariable logistic regression model, anemia defined as hemoglobin below 12 g/dL carried an adjusted odds ratio of 2.07 (95 percent confidence interval 1.20 to 3.58; p = 0.009), acidosis defined as pH below 7.35 carried an odds ratio of 2.26 (95 percent CI 1.09 to 4.95; p = 0.033), and hypobicarbonatemia below 22 mmol/L carried an odds ratio of 2.79 (95 percent CI 1.42 to 5.69; p = 0.004). Older age and out-of-hospital cardiac arrest also remained independently associated with mortality. Elevated lactate above 2.2 mmol/L showed only a borderline association (odds ratio 1.85, 95 percent CI 0.93 to 3.77; p = 0.08), and hyperglycemia was similarly borderline (odds ratio 1.72, p = 0.08). A nominal borderline association for hypercalcemia (odds ratio 3.83, p = 0.08) rested on only 12 patients and had a statistically unstable confidence interval of 0.82 to 17.14, so the authors characterized it as hypothesis-generating. Multivariable analysis revealed no significant associations between other electrolyte imbalances and mortality in this cohort.

The model performed well statistically. It showed good discrimination, with a c-statistic of 0.838 (95 percent CI 0.795 to 0.880) and a bootstrap optimism-corrected value of 0.810, along with adequate calibration (Hosmer-Lemeshow p = 0.94; Nagelkerke R2 = 0.36). Collinearity diagnostics raised no concerns: all variance inflation factors were 1.4 or lower. The only strong correlation among the continuous acid-base variables was between bicarbonate and base excess (r = 0.98), which is why base excess was not carried into the model. Sodium, potassium and chloride had shown no univariate association with mortality and were likewise excluded, while sex was not associated with mortality in the univariate analysis (p = 0.066).

Because dichotomizing continuous variables discards information and assumes an abrupt change in risk at a cut-off, the team supplemented the categorical model with analyses on the continuous scale and with restricted cubic splines using four knots. Analyzed continuously, all principal parameters were strongly associated with mortality: pH carried an odds ratio of 1.89 per 0.1-unit decrease (95 percent CI 1.62 to 2.22), bicarbonate 1.23 per 1 mmol/L decrease (1.18 to 1.30), lactate 1.33 per 1 mmol/L increase (1.26 to 1.41), hemoglobin 1.19 per 1 g/dL decrease (1.10 to 1.29) and glucose 1.46 per 50 mg/dL increase (1.30 to 1.64), all p < 0.001. The splines demonstrated significant non-linearity for pH, lactate, glucose, calcium and base excess, whereas bicarbonate and hemoglobin were consistent with linear associations. The lactate case is instructive: dichotomized at 2.2 mmol/L, a value close to the cohort median of 2.6 mmol/L, the binary variable lumps mildly and markedly elevated patients together, which the authors argue explains the borderline adjusted result rather than an absence of a prognostic effect.

Stratified analyses sharpened the picture. In-hospital mortality rose monotonically across pH strata: 6.5 percent above pH 7.45, 7.3 percent at 7.35 to 7.45, 18.3 percent at 7.2 to 7.35, 42.9 percent at 7.0 to 7.2 and 65.5 percent below 7.0, with a test for trend p < 0.001. The authors note that the steep increase below pH 7.2, and again below 7.0, indicates that severe acidosis carries substantially higher risk than the mild, often reversible acidosis seen with some acute presentations. Glucose showed a U-shaped relationship instead: mortality was 40 percent below 70 mg/dL, fell to 6.3 percent at 70 to 100 mg/dL and rose progressively to 46.2 percent above 272 mg/dL (global p < 0.001; quadratic p = 0.006). The authors caution that the cut-offs were chosen for clinical applicability and are not validated prognostic thresholds.

Sensitivity analyses tested the robustness of these findings. Of the 58 patients who died in the resuscitation room, 13 had a complete initial blood gas panel and otherwise met the inclusion criteria; re-including them in an expanded cohort of 555 patients with 123 deaths (22.2 percent) did not change the conclusions. Out-of-hospital cardiac arrest (odds ratio 4.05, p < 0.001), low bicarbonate (odds ratio 2.47, p = 0.007), low hemoglobin (odds ratio 1.97, p = 0.011) and age (odds ratio 1.03, p < 0.001) remained independently associated with mortality, while the association for low pH was attenuated to borderline significance (odds ratio 1.87, p = 0.081). The continuous and spline analyses reproduced every principal association in the primary cohort, indicating that the findings do not hinge on the specific thresholds chosen.

The results align with prior work. Georgiev and colleagues, studying 1,536 non-trauma critically ill patients, identified pH as the strongest single point-of-care marker of resuscitation-room mortality (AUC 0.81), with a substantial mortality increase below pH 7.2 and a further increase below 7.0; the present study’s optimism-corrected c-statistic of 0.81 matches that figure, which the authors describe as external corroboration across cohorts differing in center, period and endpoint. The U-shaped glucose-mortality relationship reproduces the findings of Bernhard et al., who derived the same glucose breakpoints in a comparably sized German cohort. The bicarbonate result is consistent with Wigger et al., who identified low bicarbonate as an independent predictor of short-term mortality in ICU patients with cardiogenic shock, although the present cohort is more heterogeneous and studied at an earlier stage of admission.

The authors are explicit about the study’s limits. The retrospective, single-center design restricts generalizability and precludes any causal inference between initial blood gas derangements and mortality. The exclusion rate of 31.4 percent among screened patients, largely due to missing data or early inter-hospital transfers, introduces selection bias, and excluding resuscitation-room deaths creates a survivorship bias that underestimates the true mortality burden of this population; the primary analysis should therefore be read as estimating prognostic associations among patients who survive the initial resuscitation-room phase. The blood gas sample type—arterial, venous or capillary—could not be reliably reconstructed at the patient level, precluding a stratified analysis. Only the initial admission measurement was analyzed, some strata such as the glucose extremes contained few patients, and residual confounding remains possible despite multivariable adjustment. Anemia may also mark chronic disease burden rather than directly contributing to death, and low bicarbonate may reflect several distinct underlying processes. With those caveats, the authors conclude that these readily available point-of-care parameters may support early risk stratification, and that prospective multicenter studies are needed to validate the findings and evaluate their utility in guiding early targeted interventions.

Subject of Research: Prognostic value of initial point-of-care blood gas parameters for in-hospital mortality in critically ill non-trauma emergency department patients

Article Title: Prognostic value of early point-of-care blood gas analysis in critically ill non-trauma patients

Article References: Maier-Stocker, C., Läng, R., Zinsser-Krys, J., Hanses, F., Zimmermann, M., & Hupf, J. (2026). Prognostic value of early point-of-care blood gas analysis in critically ill non-trauma patients. BMC Emergency Medicine, 26(1), Article 283. https://doi.org/10.1186/s12873-026-01822-2

Image Credits: AI Generated

DOI: 10.1186/s12873-026-01822-2

Keywords: blood gas analysis, point-of-care testing, emergency department, resuscitation room, in-hospital mortality, acidosis, anemia, bicarbonate, lactate, risk stratification, retrospective cohort study, critical illness