Study design
We conducted a post hoc analysis using the EUROBACT-2 international cohort study database. The original database contained data on 2600 adult patients from 333 ICUs across 52 countries collected between June 2019 and January 202117. For this study, we accessed data regarding patients who received antimicrobial therapy within the first 48 h from initial blood culturing, resulting in a dataset of 2406 adult patients from 328 ICUs across 52 countries. A list of participating countries is provided in Supplementary Table S1. All methods were performed in accordance with the relevant guidelines and regulations.
Ethics approval
The EUROBACT-2 study (ClinicalTrials.gov registration NCT03937245) was granted an initial ethical approval as a low-risk research project with waiver of individual informed consent by the Human Research Ethics Committee of the Royal Brisbane & Women’s Hospital, Queensland, Australia (LNR/2019/QRBW/48,376). Each study site then obtained ethical and governance approvals according to national and/or local regulations17.
Outcomes
The primary study outcomes were the proportions of patients receiving ECAT or empiric antibiotic monotherapy (EMT) for the treatment of HA-BSI. Odds ratios (ORs) with corresponding 95% credible intervals (95% CrIs) were estimated for variables that may have influenced the use of ECAT. The secondary outcomes were the number of antibiotic agents used in patients receiving ECAT and the frequencies of different antibiotic combinations used. No patient outcomes were analysed.
Definitions
The EUROBACT-2 study included adult (≥ 18 years old) patients with HA-BSI treated in the ICU17. HA-BSI was defined as a positive blood culture sample collected more than 48 h after hospital admission17. Sepsis-3 definitions for sepsis and septic shock were used6,17. A detailed explanation of definitions used in the study, including a flow-chart visualising the patient inclusion process (Supplementary Fig. S1), is available in ”Definitions”, Supplementary File 1.
Antibiotic therapy was classified as empiric when the following criteria were met: the primary indication was reported as empiric treatment for HA-BSI (i.e. not targeted therapy, de-escalation on the basis of antibiotic susceptibility testing, or treatment of other infections) by physicians participating in EUROBACT-2; treatment was initiated within 48 h after initial blood culture sampling. If empiric antibiotic therapy was started before initial blood culture sampling, it had to have continued during the first 48 h after blood culture sampling. Antifungal and antiviral agents were not included in the definition of empiric antibiotic therapy.
ECAT was defined as empiric therapy for HA-BSI using two or more antibiotic agents of different classes (e.g., beta-lactam plus aminoglycoside). EMT referred to the use of a single antibiotic agent. Patients receiving multiple antibiotic agents simultaneously, with only one agent used specifically for empiric treatment of HA-BSI, were classified in the EMT group. For example, a patient who received a broad-spectrum beta-lactam for empiric HA-BSI treatment while simultaneously receiving vancomycin as part of prior targeted therapy was included in the EMT group.
Immune deficiency was defined as the presence of any of the following: malignant tumours regardless of the presence of metastases, haematological malignancy, other solid tumours, previous organ transplantation, treatment with high-dose steroids, or other immunosuppression. The frequencies of the aforementioned diagnoses in the study population are presented in Supplementary Table S2.
Statistical analysis
A detailed description of the statistical methods used in the study is provided in ”Statistical Analysis”, Supplementary File 1. Patients were stratified into groups based on having received EMT or ECAT, and descriptive statistics tables were generated. Continuous variables were presented as medians with corresponding interquartile ranges. Categorical values were presented as counts and percentages.
A multilevel logistic regression model was used, with patients nested within ICUs, which were further nested within countries. Such clustering leads to statistical correlation and reduces the effective sample size18. Multilevel modelling accounts for this correlation, ensuring accurate estimation of uncertainty (e.g., 95% CrIs) around measures of association (e.g., ORs). From a clinical perspective, the existence of patient correlation within ICUs and countries is valuable and may reflect differences in therapeutic traditions19.
Markov-chain Monte Carlo (MCMC) estimation was used to assess the effects of patient and institutional factors on the odds of receiving ECAT. ORs with corresponding 95% CrIs were calculated by exponentiating the regression coefficients. The intra-class correlation coefficient (ICC), which measures the proportion of total variance attributable to higher-level clustering (e.g., ICUs or countries), was calculated18. Regression models containing no fixed effects, patient variables only, as well as patient and ICU variables, were compared by calculating the proportional change in variance (PCV). This approach helped distinguish whether variations in ECAT use were caused primarily by differences in patient cohorts or by contextual factors like therapeutic traditions20.
Variable selection
As the goal of regression modelling was exploratory in nature (as opposed to the construction of a prediction model), variables available in the EUROBACT-2 database were chosen based on their clinical relevance as determined by the study authors. Variables were excluded from the final regression analysis due to collinearity, e.g. vasopressor support was not included in the model as it is already included in the definition of septic shock.
Subgroup analyses
Subgroup regression analyses were performed based on the profile of empiric antibiotic coverage: gram-negative coverage; gram-positive coverage; anaerobe coverage; coverage for atypical pathogens.
Sensitivity analysis
A proportion of patients included in the EUROBACT-2 study received simultaneous treatment for infections other than HA-BSI. A sensitivity analysis including patients who received ECAT or EMT exclusively for the treatment of HA-BSI was performed. This was done due to the possibility that treatment of other infections, initiated prior to the onset of HA-BSI, may have affected the choice or reporting of ECAT and EMT. For example, for a patient with a new HA-BSI where ECAT with beta-lactams and aminoglycosides is indicated, and where treatment with a beta-lactam for another infection is already ongoing, the reporting physician may have reported only the initiation of EMT with an aminoglycoside (this being the only new treatment started), as opposed to registering ECAT with a beta-lactam and aminoglycoside combination (this being the actual agents exerting antimicrobial effects in vivo).
Missing data
A complete case analysis was performed due to a very low proportion of missing data.
Software
Statistical analysis was performed using the R statistical software environment (v4.4.2; R Foundation for Statistical Computing, Vienna, Austria) and MlwiN (v3.06; Centre for Multilevel Modelling, University of Bristol)21,22,23,24,25,26,27,28,29,30.
