Outcomes adjusted for confounders.

This table shows how the obtained results are affected by commonly criticized confounders. To adjust for differences in antibiotic usage and heterogeneity, we predicted new estimates for univariate with the function predict() as implemented in the package metafor in R. Here, antibiotic heterogeneity is defined with the antibiotic heterogeneity index, with n = number of employed antibiotics and ai = usage of antibiotic a/total antibiotic usage.To adjust for antibiotic heterogeneity and consumption, we predicted the estimates for the hypothetical case that the ratio of daily defined doses and antibiotic heterogeneity indices is 1, i.e. exactly equal antibiotic consumption and heterogeneity in both study arms. For predicting the difference when controlling for hospital import, we predicted estimates for the hypothetical case that all studies only report isolates of strains that the respective patients were neither colonized nor infected with at admission.