Matt Harmon

51 Chapter three cytometric bead arrays (BD Bioscience, San Jose, CA) using FACSCalibur (Becton Dickenson, Franklin Lakes, NJ, USA). Angiopoietin-1 and angiopoietin-2 (R&D systems, Abingdon, UK) were measured by Luminex multiplex assay using BioPlex 200 (BioRad, Hercules, CA). Normal biomarker values were acquired from EDTA plasma from 27 age- and gender-matched healthy volunteers, from whom written informed consent was obtained. The lower limits of detection for the immune assays were: 0.9 pg/mL for TNF- α , 1.3 pg/mL for IL-1 β , 0.9 pg/mL for IL-6, 1.3 pg/mL for IL-8, 0.8 pg/ mL for IL-10, 0.7 pg/mL for IL-13, 3.1 pg/mL for soluble E-selectin, 6.3 pg/mL for soluble ICAM-1, 4.0 pg/mL for fractalkine, 0.2 pg/mL for angiopoietin-1 and 1.8 pg/mL for angiopoietin-2. Whole blood stimulations In a random subset of 15 patients, whole blood was stimulated ex vivo with LPS on day 1 of ICU admission, as previously described. 20 Heparin-anticoagulated blood was stimulated for 3h at 37°C in pyrogen-free RPMI 1640 (Life Technologies, Bleiswijk, the Netherlands) with or without 100ng/mL ultrapure LPS (from Escherichia coli  0111:B4; InvivoGen, Toulouse, France). TNF- α and IL-1 β were measured in supernatants using a cytometric bead array assay (BD Biosciences, San Jose, California). Cytokine release was calculated as the difference in cytokine levels in samples incubated with and without LPS. The medical ethical committee of the Academic Medical Center in Amsterdam gave ethical approval for the conduction of the study (no.NL34294.018.10). Written informed consent was obtained from all patients, or their legal representative, and from healthy volunteers. Statistical analysis All analyses were performed in R (version 3.1.1). Student’s t-test or Wilcoxon rank-sum test, and chi-square test were used to compare groups. To study factors independently associated with developing hypothermia, we performed multivariable logistic regression. Pre-ICU-admission patient characteristics that were deemed relevant or were associated with hypothermia in univariate analysis ( P <0.2) were included in the model. A backward selection procedure using the Akaike information criterion (AIC) including 1000 bootstrap replicates was applied (R-package “rms”) to identify risk factors for hypothermia. Age was forced into the model since we considered age an important confounder for all factors incorporated in the model as well as being associated with an altered temperature response. 4,21