Prognostic and pathogenetic value of combining clinical and biochemical indices in patients with acute lung injury

Abstract

Background: No single clinical or biologic marker reliably predicts clinical outcomes in acute lung injury (ALI)/ARDS. We hypothesized that a combination of biologic and clinical markers would be superior to either biomarkers or clinical factors alone in predicting ALI/ARDS mortality and would provide insight into the pathogenesis of clinical ALI/ARDS.

Methods: Eight biologic markers that reflect endothelial and epithelial injury, inflammation, and coagulation (von Willebrand factor antigen, surfactant protein D [SP-D]), tumor necrosis factor receptor-1, interleukin [IL]-6, IL-8, intercellular adhesion molecule-1, protein C, plasminogen activator inhibitor-1) were measured in baseline plasma from 549 patients in the ARDSNet trial of low vs high positive end-expiratory pressure. Mortality was modeled with multivariable logistic regression. Predictors were selected using backward elimination. Comparisons between candidate models were based on the receiver operating characteristics (ROC) and tests of integrated discrimination improvement.

Results: Clinical predictors (Acute Physiology And Chronic Health Evaluation III [APACHE III], organ failures, age, underlying cause, alveolar-arterial oxygen gradient, plateau pressure) predicted mortality with an area under the ROC curve (AUC) of 0.82; a combination of eight biomarkers and the clinical predictors had an AUC of 0.85. The best performing biomarkers were the neutrophil chemotactic factor, IL-8, and SP-D, a product of alveolar type 2 cells, supporting the concept that acute inflammation and alveolar epithelial injury are important pathogenetic pathways in human ALI/ARDS.

Conclusions: A combination of biomarkers and clinical predictors is superior to clinical predictors or biomarkers alone for predicting mortality in ALI/ARDS and may be useful for stratifying patients in clinical trials. From a pathogenesis perspective, the degree of acute inflammation and alveolar epithelial injury are highly associated with the outcome of human ALI/ARDS.

Figure 1.

Receiver operator characteristic curves for predictive models of mortality that include the Acute Physiology And Chronic Health Evaluation (APACHE) III score in 528 patients with acute lung injury (ALI)/ARDS. The full model includes all six clinical predictors and all eight biomarker variables. The reduced model includes APACHE III score, age, surfactant protein D (SP-D), and interleukin (IL)-8. AUC = area under the receiver operator curve.

Figure 2.

Receiver operator characteristic curves for predictive models of mortality that include the APACHE III score in 528 patients with ALI/ARDS. The full model includes all six clinical predictors and all eight biomarker variables. The reduced model includes age, number of organ failures, alveolar-arterial oxygen difference, SP-D, IL-8, plasminogen activator receptor (PAI)-1, and tumor necrosis factor receptor (TNFR) 1. See Figure 1 legend for expansion of other abbreviations.

Figure 3.

Calibration curve for the reduced model that includes age, number of organ failures, alveolar-arterial oxygen difference, SP-D, IL-8, PAI-1, and TNFR1. The gray line represents perfect calibration. Where the observed proportion of death (black line) is above the gray line, there are more deaths than predicted by the model. The model performs well across the spectrum of predicted probabilities of mortality. See Figure 1 and 2 legends for expansion of abbreviations.

Figure 4.

Predicted probability of death for each clinical variable in the reduced model without APACHE III when all other variables are fixed at their median values. Median values are: age, 50 years; number of organ failures at enrollment, 0; arterial-alveolar oxygen difference, 286 mm Hg. Dotted lines in each plot represent the 95% CI. See Figure 1 legend for expansion of abbreviation.

Figure 5.

Predicted probability of death for each biomarker variable in the reduced model without APACHE III when all other variables are fixed at their median values. Median values are: IL-8, 40.4 pg/mL; SPD, 99.0 ng/mL; PAI, 61.3 ng/mL; TNFR1, 4,283 pg/mL. Dotted lines in each plot represent the 95% CI. See Figure 1 and 2 legends for expansion of abbreviations.