A 2-Biomarker Model Augments Clinical Prediction of Mortality in Melioidosis

Abstract

Background: Melioidosis, infection caused by Burkholderia pseudomallei, is a common cause of sepsis with high associated mortality in Southeast Asia. Identification of patients at high likelihood of clinical deterioration is important for guiding decisions about resource allocation and management. We sought to develop a biomarker-based model for 28-day mortality prediction in melioidosis. Methods: In a derivation set (N=113) of prospectively enrolled, hospitalized Thai patients with melioidosis, we measured concentrations of interferon-γ, interleukin-1β, interleukin-6, interleukin-8, interleukin-10, tumor necrosis factor-E', granulocyte-colony stimulating factor, and interleukin-17A. We used least absolute shrinkage and selection operator (LASSO) regression to identify a subset of predictive biomarkers and performed logistic regression and receiver operating characteristic curve analysis to evaluate biomarker-based prediction of 28-day mortality compared with clinical variables. We repeated select analyses in an internal validation set (N=78) and in a prospectively enrolled external validation set (N=161) of hospitalized adults with melioidosis. Results: All 8 cytokines were positively associated with 28-day mortality. Of these, interleukin-6 and interleukin-8 were selected by LASSO regression. A model consisting of interleukin-6, interleukin-8, and clinical variables significantly improved 28-day mortality prediction over a model of only clinical variables [AUC (95% confidence interval [CI]): 0.86 (.79-.92) vs 0.78 (.69-.87); P=.01]. In both the internal validation set (0.91 [0.84-0.97]) and the external validation set (0.81 [0.74-0.88]), the combined model including biomarkers significantly improved 28-day mortality prediction over a model limited to clinical variables. Conclusions: A 2-biomarker model augments clinical prediction of 28-day mortality in melioidosis.