Fig. 2

Effects of antimicrobial therapy and of surgical source control on 28-day mortality. Effects were tested in a logistic hierarchical linear model with a random intercept adjusting for covariates. Risk-adjusted mortality estimates were obtained as predictive margins that were calculated for the average of continuous variables and for the most common category of categorical variables. No. of patients gives the number of cases with complete data compared to the total number of patients suitable for the respective analysis. * marks the p-value of the overall test of significance for the categorical variables on timing conducted by a likelihood-ratio test, while the other p-values give the results of tests of single categories against the reference category. Models adjusted for the following covariates: age and gender, origin of infection, location of the patient at the onset of sepsis, focus of infection, microbiological confirmation of infection, study phase (trial vs. surveillance phase), and group the hospital was randomized to (intervention vs. control)