Identifying heterogeneity of treatment effect for antibiotic duration in bloodstream infection: an exploratory post-hoc analysis of the BALANCE randomised clinical trial.

Background
The BALANCE trial demonstrated non-inferiority of 7 (vs 14) day antibiotic durations in patients with uncomplicated non-S. aureus/lugdunensis bacterial bloodstream infections (BSI). However, there may be patient subgroups who benefit from longer durations. We aimed to evaluate if bedside clinical decision rules could identify these subgroups.
Methods
In this post-hoc analysis of the multicentre, randomised BALANCE trial (October 17, 2014-May 5, 2023), we applied three clinical decision rules to investigate heterogeneity of treatment effect in 7-day vs 14-day antibiotic durations on 90-day all-cause mortality. We used the rules to categorize patients in BALANCE into different risk groups and calculated the unadjusted absolute risk difference (RD) for 90-day mortality in patients receiving 7- vs 14-day antibiotics within each risk group. Statistical significance was tested using an interaction test. The BALANCE trial is registered with ClinicalTrials.gov (NCT03005145).
Findings
3581 patients were included. All three rules predicted mortality risk, but none identified statistically significant effect modification: (a) static rule (low-risk: RD -0.58, 95% CI -8.91 to 7.73; moderate-risk: RD -.01, 95% CI -3.86 to 1.83; high-risk: RD -2.65, 95% CI -7.12 to 1.81; p = 0.74); (b) dynamic rule (met rule on day 7: RD -2.18, 95% CI -4.81 to 0.45; did not meet rule: RD 1.75, 95% CI -3.89 to 7.40; p = 0.16); and (c) early clinical failure criteria (score<2: RD -2.38, 95% CI -5.0 to 0.23; score ≥2: RD -0.65, 95% CI -5.06 to 3.77; p = 0.24). Results were consistent across sensitivity analyses including imputation for missing data and restricting analyses to gram-negative BSI.
Interpretation
The decision rules included in our analyses did not identify a subgroup of patients within BALANCE that would benefit from 14 (vs 7) days of treatment. 7-day treatment duration is sufficient for most patients with uncomplicated non-S. aureus/lugdunensis BSI. Future research could explore data-driven machine-learning approaches to identify comprehensive combinations of patient characteristics that may guide individualised duration of antibiotic therapy.
Funding
The BALANCE trial was funded by the Canadian Institutes of Health Research, Health Research Council of New Zealand, Australian National Medical Research Council, Physicians Services Incorporated Ontario and Ontario Ministry of Health and Long-term Care Innovation Fund. SWXO conducted this study as part of his PhD studies, with funding from: the Emerging & Pandemic Infections Consortium (University of Toronto, Canada); Connaught International Scholarship (University of Toronto, Canada); the Queen Elizabeth II Graduate Scholarship in Science and Technology (QEII-GSST; Government of Ontario, Canada); and the Melbourne Research Scholarship (University of Melbourne, Australia). VML is supported by Clinical Research Scholar-Junior 2 program (FRQ-S).