IL-1R2-based biomarker models predict melioidosis mortality independent of clinical data

IntroductionMelioidosis is an often-fatal tropical infectious disease caused by the Gram-negative bacillus Burkholderia pseudomallei, but few studies have identified promising biomarker candidates to predict outcome. MethodsIn 78 prospectively enrolled patients hospitalized with melioidosis, six candidate protein biomarkers, identified from the literature, were measured in plasma at enrollment. A multi-biomarker model was developed using least absolute shrinkage and selection operator (LASSO) regression, and mortality discrimination was compared to a clinical variable model by receiver operating characteristic curve analysis. Mortality prediction was confirmed in an external validation set of 191 prospectively enrolled patients hospitalized with melioidosis. ResultsLASSO regression selected IL-1R2 and soluble triggering receptor on myeloid cells 1 (sTREM-1) for inclusion in the candidate biomarker model. The areas under the receiver operating characteristic curve (AUC) for mortality discrimination for the IL-1R2 + sTREM-1 model (AUC 0.81, 95% CI 0.72-0.91) as well as for an IL-1R2-only model (AUC 0.78, 95% CI 0.68-0.88) were higher than for a model based on a modified Sequential Organ Failure Assessment (SOFA) score (AUC 0.69, 95% CI 0.56-0.81, p < 0.01, p = 0.03, respectively). In the external validation set, the IL-1R2 + sTREM-1 model (AUC 0.86, 95% CI 0.81-0.92) had superior 28-day mortality discrimination compared to a modified SOFA model (AUC 0.80, 95% CI 0.74-0.86, p < 0.01) and was similar to a model containing IL-1R2 alone (AUC 0.82, 95% CI 0.76-0.88, p = 0.33). ConclusionBiomarker models containing IL-1R2 had improved 28-day mortality prediction compared to clinical variable models in melioidosis and may be targets for future, rapid test development.

Automated summary

IL-1R2 and sTREM-1 are promising biomarkers for predicting mortality in melioidosis, and biomarker models containing IL-1R2 have improved prediction compared to clinical variable models.