Matt Harmon

Chapter four 79 Briefly, raw scans were pre-processed by means of the robust multi-average (RMA) method, normalized (quantile), summarized by median polish and log2 transformed using the affy method. 20 Non-experimental chip effects were assessed and corrected by means of the combat method in the surrogate variable analysis R package. 21 Comparisons between groups was done using multi-variate linear models, including age and gender as covariates, implemented in the limma method. 22 Benjamini-Hochberg (BH) adjusted p-values < 0.05 defined genome- wide significance. To assess the association with canonical signaling pathways we used Ingenuity Pathway Analysis software (Qiagen Bioinformatics). Fisher exact test BH-adjusted p-values < 0.05 demarcated significance. Human species and Ingenuity gene knowledgebase were specified. All other parameters were default. Statistical analysis All data were analyzed using R studio (version 3.2.2, R Core Team 2013, Vienna, Austria). In this study we performed two analyses. Firstly, we compared genomic profiles between septic hypothermic and febrile patients without correcting for disease severity. Subsequently, in order to determine whether hypothermia was associated with a specific genomic profile irrespective of severity of disease, hypothermic patients were 1:1 matched to fever patients using their APACHE IV scores and presence of shock. Data is presented as numbers (percentages), parametric data as mean ± SD and non-parametric data as median and 25 th – 75 th percentages; Q1-Q3). Data distribution was assessed by the Kolmogorov-Smirnov test. Mann-Whitney U or a Kruskal-Wallis test was used to analyze continuous nonparametric data, whereas continuous parametric data were analyzed using Student’s t-test or analysis of variance (two-sided analysis of variance). All categorical data were analyzed using a chi-square or Fisher exact test. A p-value less than 0.05 was considered to be of statistical significance for clinical data. Matching was done using “optimal matching” with R-package “MatchIt” (caliper 0.35 standard deviations of the logit). Results Patients The selection of study patients can be seen in figure 1. Out of a total of 579 sepsis admissions, 168 patients were included in the microarray analysis. Table 1 shows the baseline characteristics of these patients arranged by temperature group (baseline characteristics of patients that were normothermic (n=75) or both