A Two-Biomarker Model Predicts Mortality in the Critically Ill with Sepsis

Abstract

Rationale: Improving the prospective identification of patients with systemic inflammatory response syndrome (SIRS) and sepsis at low risk for organ dysfunction and death is a major clinical challenge.

Objectives: To develop and validate a multibiomarker-based prediction model for 28-day mortality in critically ill patients with SIRS and sepsis.

Methods: A derivation cohort (n = 888) and internal test cohort (n = 278) were taken from a prospective study of critically ill intensive care unit (ICU) patients meeting two of four SIRS criteria at an academic medical center for whom plasma was obtained within 24 hours. The validation cohort (n = 759) was taken from a prospective cohort enrolled at another academic medical center ICU for whom plasma was obtained within 48 hours. We measured concentrations of angiopoietin-1, angiopoietin-2, IL-6, IL-8, soluble tumor necrosis factor receptor-1, soluble vascular cell adhesion molecule-1, granulocyte colony-stimulating factor, and soluble Fas.

Measurements and main results: We identified a two-biomarker model in the derivation cohort that predicted mortality (area under the receiver operator characteristic curve [AUC], 0.79; 95% confidence interval [CI], 0.74-0.83). It performed well in the internal test cohort (AUC, 0.75; 95% CI, 0.65-0.85) and the external validation cohort (AUC, 0.77; 95% CI, 0.72-0.83). We determined a model score threshold demonstrating high negative predictive value (0.95) for death. In addition to a low risk of death, patients below this threshold had shorter ICU length of stay, lower incidence of acute kidney injury, acute respiratory distress syndrome, and need for vasopressors.

Conclusions: We have developed a simple, robust biomarker-based model that identifies patients with SIRS/sepsis at low risk for death and organ dysfunction.

Keywords: IL-8; biomarkers; sepsis; systemic inflammatory response syndrome; tumor necrosis factor receptor.

Figure 1.

Sample and statistical flow diagram for mortality prediction model. Subject numbers for the development and validation of the least absolute shrinkage and selection operator or all subsets selection logistic regression models for 28-day mortality. *Lower limit or upper limit of detection applied to samples falling below or above the range of detection. Ang = angiopoietin; BMI = body mass index; CKD = chronic kidney disease; G-CSF = granulocyte colony–stimulating factor; ICU = intensive care unit; LASSO = least absolute shrinkage and selection operator; sVCAM = soluble vascular cell adhesion molecule; TNFR = tumor necrosis factor receptor.

Figure 2.

Receiver operating curves for mortality prediction models. Area under the receiver operating characteristic curve (AUC) for mortality in the (A) derivation cohort, (B) internal test cohort, and (C) external validation cohort. AUC values are presented with 95% confidence intervals in parenthesis. The models are shown as TNFR + IL8 to represent the model of two biomarkers alone, Acute Physiology and Chronic Health Evaluation III (APC3), and TNFR + IL8 + APC3 for the combined model of biomarkers and APC3. TNFR = tumor necrosis factor receptor.