Inflammatory biomarkers in infective endocarditis: machine learning to predict mortality.

Infective endocarditis (IE) is the cardiac disease with the highest rates of mortality. New biomarkers that are able to identify patients at risk for death are required to improve patient management and outcome. This study aims to investigate if cytokines, chemokines and growth factors measured at IE diagnosis can predict mortality. Patients with definite IE, according to the Duke's modified criteria, were included. Using high-performance Luminex assay, 27 different cytokines, chemokines and growth factors were analyzed. Machine learning techniques were used for the prediction of death and subsequently creating a decision tree, in which the cytokines, chemokines and growth factors were analyzed together with C-reactive protein (CRP). Sixty-nine patients were included, 41 (59%) male, median age 54 [interquartile range (IQR) = 41-65 years] and median time between onset of the symptoms and diagnosis was 12 days (IQR = 5-30 days). The in-hospital mortality was 26% (n = 18). Proinflammatory cytokines interkeukin (IL)-15 and C-C motif chemokine ligand (CCL4) were found to predict death, adding value to CRP levels. The decision tree predicted correctly the outcome of 91% of the patients at hospital admission. The high-risk group, defined as CRP ≥ 72 mg/dL, IL-15 ≥ 5·6 fg/ml and CCL4 ≥ 6·35 fg/ml had an 88% in-hospital mortality rate, whereas the patients classified as low-risk had a mortality rate of 8% (P = < 0·001). Cytokines IL-15 and CCL4 were predictors of mortality in IE, adding prognostic value beyond that provided by CRP levels. Assessment of cytokines has potential value for clinical risk stratification and monitoring in IE patients.