Highlights
Patients randomized to 7 days had higher non‑adherence (24.0%) than those randomized to 14 days (16.6%); non‑adherence was associated with important prognostic variables.
After adjustment for non‑adherence using inverse probability of treatment weighting (IPTW) and instrumental variable analyses, effect estimates for 7 versus 14 days on 90‑day all‑cause mortality remained consistent with non‑inferiority.
Causal inference methods strengthened confidence that a 7‑day antibiotic course should be standard for uncomplicated, non‑Staphylococcus aureus bloodstream infection (BSI), while underscoring the importance of careful patient selection and reporting of adherence in trials.
Background
Bloodstream infection (BSI) is a common and potentially life‑threatening condition that frequently prompts prolonged courses of intravenous and/or oral antibiotics. Prolonged therapy increases costs, hospital stay, line‑associated complications, adverse drug events, and selection pressure for antimicrobial resistance. Accordingly, antibiotic stewardship has prioritized defining the minimum effective treatment duration that achieves clinical cure without excess mortality or relapse.
The BALANCE trial (NCT03005145) was designed as a randomized, non‑inferiority study comparing 7‑day versus 14‑day antibiotic courses in adults with uncomplicated BSI due to organisms other than Staphylococcus aureus. The primary publication showed non‑inferiority of the shorter course for 90‑day all‑cause mortality. However, non‑adherence to assigned duration occurred in approximately 20% of participants, raising the possibility of bias in the intention‑to‑treat (ITT) analysis if non‑adherence was related to prognosis or disease severity.
This post‑hoc analysis by Ong et al. (Lancet Infect Dis 2025) interrogates factors associated with non‑adherence and applies causal inference methods to estimate the effect of treatment duration on mortality while accounting for deviations from assigned therapy.
Study design and analytic approach
Population: 3,581 trial participants with complete outcome data from the BALANCE trial, excluding S. aureus bloodstream infections and focusing on uncomplicated cases.
Interventions and comparators: Randomization to antibiotic duration of 7 days versus 14 days.
Primary endpoint: 90‑day all‑cause mortality.
Adherence definitions: Two operationalized definitions were used. First, a trial‑specific daily adherence variable recorded by site research staff defining adherence as receipt of antibiotic duration within ±2 days of the assigned course. Second, a total antibiotic duration variable recording cumulative days of systemic antibiotic therapy.
Analytic methods: The authors identified predictors of non‑adherence using generalized linear mixed models. To address potential bias introduced by non‑adherence, they used two causal inference strategies: inverse probability of treatment weighting (IPTW) applied to per‑protocol and as‑treated populations, and instrumental variable (IV) analyses leveraging randomization assignment as an instrument for received duration. Both adherence definitions were used in sensitivity analyses. The same 4% non‑inferiority margin used in the primary trial was retained.
Key findings
Extent and predictors of non‑adherence
Overall non‑adherence occurred in 728 of 3,581 patients (20.3%). Non‑adherence was more common in the 7‑day arm (432/1,802; 24.0%) than the 14‑day arm (296/1,779; 16.6%). Important clinical features were associated with prolongation of therapy: higher disease severity, intra‑abdominal and skin/soft tissue infection sources, persistent fever, and persistent bacteremia. Factors associated with shortening of therapy included vascular catheter source and presence of antimicrobial resistance.
Adjusted causal estimates for 90‑day mortality
After accounting for measured confounding from non‑adherence, adjusted risk differences (7‑day minus 14‑day) for 90‑day all‑cause mortality were broadly in line with the primary ITT results and consistent with non‑inferiority:
– IPTW per‑protocol analysis (adherence definition one): risk difference −1.61% (95% CI −4.13 to 0.87).
– IPTW per‑protocol analysis (definition two): −1.91% (95% CI −4.61 to 0.93).
– IPTW as‑treated analysis: −0.75% (95% CI −3.36 to 1.86).
– Instrumental variable analysis (definition one): −2.68% (95% CI −6.65 to 1.29).
– Instrumental variable analysis (definition two): −2.80% (95% CI −5.98 to 0.37).
Across methods and adherence definitions, point estimates favored the 7‑day strategy (negative risk differences) and confidence intervals overlapped the prespecified 4% non‑inferiority margin in a manner consistent with non‑inferiority.
Safety and secondary outcomes
The post‑hoc report focuses on the primary outcome and methodologic evaluation of bias; detailed secondary safety outcomes (e.g., relapse, Clostridioides difficile infection, catheter complications) were not the central subject of this analysis and remain described in the primary trial publication.
Interpretation and methodological implications
Why non‑adherence matters
Non‑adherence in randomized trials can undermine causal inference when deviations from assigned therapy are related to prognosis — a classic source of informative censoring or treatment‑selection bias. In BALANCE, clinicians disproportionately prolonged therapy in clinically sicker patients or those with persistent signs, and shortened therapy in other subgroups. If unaddressed, such differential non‑adherence could bias per‑protocol or as‑treated comparisons away from the true randomized effect.
Strengths of the analytic approach
The authors appropriately characterized predictors of non‑adherence and applied two complementary causal inference methods. IPTW seeks to emulate a randomized comparison among those who would adhere, by weighting patients to balance measured prognostic factors associated with adherence. Instrumental variable analysis leverages randomization as an instrument to estimate causal effect in the presence of non‑adherence under assumptions (relevance, independence, exclusion restriction). That both approaches produced similar conclusions increases confidence that measured non‑adherence did not materially change the trial’s conclusion of non‑inferiority.
Limitations and residual concerns
– Post‑hoc nature: This is a secondary analysis and, by definition, hypothesis‑generating about mechanisms of bias rather than altering the randomized inference.
– Measured versus unmeasured confounding: IPTW adjusts for observed predictors of non‑adherence, but residual confounding by unmeasured severity markers or clinician judgment remains possible. IV analysis is robust to unmeasured confounding only if the instrument assumptions hold; the exclusion restriction (that randomization affects outcome only through received duration) can be challenging to verify when care behaviors differ by allocation.
– Adherence definitions: Two different adherence operationalizations were necessary; neither is perfect. The trial‑staff daily assessment may better capture intention and planned deviations, whereas total duration captures cumulative exposure but may mask timing-specific patterns (e.g., short interruptions).
– Generalizability: The trial excluded Staphylococcus aureus bacteremia, endovascular infections, and other complicated syndromes. Results therefore apply to uncomplicated BSI only; clinicians should not extrapolate to S. aureus bacteremia, endocarditis, deep‑seated infections, or immunocompromised hosts without supporting evidence.
– Outcomes beyond mortality: Important stewardship outcomes (resistance emergence, C. difficile, catheter complications, readmissions) require ongoing follow‑up and pooling of trial results to gauge broader harms/benefits of shorter therapy.
Clinical implications
This robust secondary analysis strengthens the primary BALANCE finding: for adults with uncomplicated bloodstream infection due to non‑Staphylococcus aureus organisms, a 7‑day antibiotic course is non‑inferior to 14 days for 90‑day all‑cause mortality, even after accounting for meaningful rates of non‑adherence and its prognostic correlates. For frontline clinicians and stewardship programs, the analysis supports adopting 7 days as standard for eligible patients, accompanied by clear documentation of infection source control, clinical stability, and follow‑up plans.
Clinical caveats include ensuring appropriate source control, excluding complicated infections (deep/undrained foci, prosthetic material involvement, endovascular infection), and individualizing duration for immunocompromised patients, persistent bacteremia, or recrudescence of signs.
Expert commentary
The BALANCE post‑hoc analysis exemplifies best practice for handling non‑adherence: transparent reporting of who deviated and why, and the use of causal inference tools to estimate the per‑protocol effect. Methodologists will note the value of prespecifying adherence definitions and analytic approaches in trial protocols to avoid post‑hoc multiplicity. Practically, the data align with a growing body of randomized trials supporting shorter courses for many bacterial infections, reinforcing stewardship principles without compromising survival.
Conclusion
Non‑adherence in BALANCE was common and related to clinically plausible prognostic factors, but did not materially change the trial’s primary inference after rigorous adjustment. The convergent findings from IPTW and instrumental variable analyses bolster confidence that a 7‑day course is an effective and safe standard for uncomplicated non‑Staphylococcus aureus BSI in appropriately selected adult patients. Implementation should be accompanied by clear eligibility criteria, attention to source control, and monitoring for complications.
Suggested areas for future research
– Prospective trials that prespecify adherence definitions and incorporate pragmatic, real‑world adherence monitoring.
– Trials and observational studies to define safe minimum durations in immunocompromised hosts and S. aureus bacteremia.
– Integrated stewardship initiatives measuring ecological outcomes (resistance, C. difficile) associated with reduced antibiotic days at population level.
– Qualitative work to better understand clinician decision drivers for extending or shortening therapy in BSI.
Funding and trial registration
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. ClinicalTrials.gov identifier: NCT03005145.
References
1) Ong SWX, Pinto R, Mahar RK, Rishu A, Davis JS, Fowler RA, Tong SYC, Daneman N; BALANCE trial consortium. Accounting for non‑adherence to assigned antibiotic treatment duration for bloodstream infection (BALANCE): a post‑hoc analysis of a randomised clinical trial. Lancet Infect Dis. 2025 Nov 11:S1473‑3099(25)00592‑4. doi: 10.1016/S1473‑3099(25)00592‑4. Epub ahead of print. PMID: 41237792 IF: 31.0 Q1 .
2) Singer M, Deutschman CS, Seymour CW, et al. The Third International Consensus Definitions for Sepsis and Septic Shock (Sepsis‑3). JAMA. 2016;315(8):801‑810.