Skip to main content

Foundational Trials

BLING III

Image: Shutterstock / ChaNaWiT

Publication

  • Title: Continuous vs Intermittent β-Lactam Antibiotic Infusions in Critically Ill Patients With Sepsis: The BLING III Randomized Clinical Trial
  • Acronym: BLING III
  • Year: 2024
  • Journal published in: JAMA
  • Citation: Dulhunty JM, Brett SJ, De Waele JJ, et al; BLING III Study Investigators and the ANZICS Clinical Trials Group. Continuous vs intermittent β-lactam antibiotic infusions in critically ill patients with sepsis: the BLING III randomized clinical trial. JAMA. 2024;332(8):629-637.

Context & Rationale

  • Background
    β-lactam antibiotics exhibit time-dependent killing; in critical illness, large pharmacokinetic variability (augmented renal clearance, altered volume of distribution, and organ dysfunction) can lead to inadequate free-drug exposure above the minimum inhibitory concentration when using conventional intermittent dosing.1
    Prior randomised trials and meta-analyses suggested possible improvements in clinical cure and (in some syntheses) mortality with prolonged/continuous infusion, but were limited by small sample sizes, heterogeneous patient populations/antibiotics, and variable methodological quality—leaving practice highly variable and recommendations weak/conditional.2
  • Research Question/Hypothesis
    In adult ICU patients with sepsis prescribed piperacillin-tazobactam or meropenem, does administration by continuous infusion (vs intermittent infusion) reduce all-cause 90-day mortality?
  • Why This Matters
    β-lactams are foundational agents in ICU sepsis care; even a small absolute mortality effect could be clinically important at population scale, but continuous infusion has non-trivial implementation costs (infusion pumps/lines, nursing workload, drug stability), and may plausibly alter antimicrobial stewardship and resistance ecology—necessitating definitive trial evidence.

Design & Methods

  • Research Question: Among critically ill adults with sepsis treated with piperacillin-tazobactam or meropenem, does continuous infusion (vs intermittent infusion) improve 90-day survival?
  • Study Type: International, multicentre, pragmatic, phase 3, open-label, parallel-group randomised clinical trial (1:1); 104 adult ICUs in Australia, Belgium, France, Malaysia, New Zealand, Sweden, and the UK.
  • Population:
    • Core setting: Adult ICU patients with suspected/documented infection and organ dysfunction.
    • Key inclusion (examples of operational eligibility): suspected/documented infection; expected ICU stay beyond “the day after tomorrow” (operationally ≥48 hours); commenced on piperacillin-tazobactam or meropenem for the current episode; and ≥1 organ dysfunction criterion within the preceding 24 hours (e.g., MAP <60 mmHg for ≥1 hour; vasopressors >4 hours; respiratory support including high-flow nasal oxygen/CPAP/BiPAP/invasive ventilation for ≥1 hour; serum creatinine >220 µmol/L).1
    • Key exclusions: age <18 years; receipt of piperacillin-tazobactam/meropenem for >24 hours for the current episode; pregnancy; known allergy to study antibiotics/penicillin; kidney replacement therapy at randomisation; limitations of advanced life support over the next 48 hours; death imminent/inevitable; prior enrolment in BLING III.1
  • Intervention:
    • Continuous infusion strategy: clinician-determined total 24-hour dose of piperacillin-tazobactam or meropenem delivered by continuous infusion (target “continuous over 24 hours”); all participants received at least 1 infusion dose prior to randomised treatment.
    • Loading/transition rules: switching between piperacillin-tazobactam and meropenem was permitted; a loading dose by intermittent infusion was required when switching for participants assigned to continuous infusion.
    • Duration: continued for the remainder of the prescribed treatment course or until ICU discharge (whichever occurred first).
  • Comparison:
    • Intermittent infusion strategy: clinician-determined dosing schedule with administration by intermittent infusion (nominally over 30 minutes).
    • Flexibility: switching between piperacillin-tazobactam and meropenem permitted within the treatment course (consistent with pragmatic ICU practice).
    • Duration: for the remainder of the prescribed course or until ICU discharge.
  • Blinding: Open-label (no blinding of treating clinicians or bedside staff); primary endpoint was objective (mortality), but secondary endpoints (e.g., “clinical cure”) could be influenced by care decisions.
  • Statistics: A total of 7000 patients were required to detect a 3.5% absolute reduction in 90-day mortality (from 27.5% to 24.0%) with 90% power at a 2-sided 5% significance level (allowing 5% withdrawal/loss). Primary analysis was modified intention-to-treat (participants providing consent/approval), using logistic regression with a site random effect; a prespecified adjusted analysis included sex, APACHE II score, admission source, and β-lactam administered prior to randomisation; multiplicity control for secondary/tertiary outcomes used Holm-Bonferroni adjustment.3
  • Follow-Up Period: Primary endpoint at day 90; key secondary outcomes assessed to day 14 (e.g., clinical cure; new acquisition/colonisation/infection with multidrug-resistant organisms or Clostridioides difficile), and tertiary “days alive and free” outcomes assessed to day 90.

Key Results

This trial was not stopped early. Recruitment exceeded the planned sample size to facilitate a prespecified pharmacokinetic/pharmacodynamic substudy after reaching 7000 participants.

Outcome Continuous infusion Intermittent infusion Effect p value / 95% CI Notes
All-cause mortality (day 90) 864/3474 (24.9%) 939/3507 (26.8%) OR 0.91 95% CI 0.81 to 1.01; P=0.08 Absolute difference −1.9% (95% CI −4.9 to 1.1); primary analysis (unadjusted model).
All-cause mortality (day 90; prespecified adjusted analysis) 864/3474 (24.9%) 939/3507 (26.8%) OR 0.89 95% CI 0.79 to 0.99; P=0.04 Adjusted for sex, APACHE II, admission source, and β-lactam prior to randomisation; absolute difference −2.2% (95% CI −5.5 to 1.1).
Clinical cure (day 14) 1930/3467 (55.7%) 1744/3491 (50.0%) OR 1.26 95% CI 1.15 to 1.38; P<0.001 Absolute difference 5.7% (95% CI 2.4 to 9.1); P values for secondary outcomes adjusted by Holm-Bonferroni.
New acquisition/colonisation/infection with multidrug-resistant organisms or C difficile (day 14) 253/3498 (7.2%) 266/3533 (7.5%) OR 0.96 95% CI 0.80 to 1.15; P=0.65 Absolute difference −0.3% (95% CI −1.9 to 1.4); excluding C difficile: 215/3498 (6.1%) vs 232/3533 (6.6%), OR 0.92 (95% CI 0.76 to 1.12), P=0.41.
All-cause ICU mortality 595/3474 (17.1%) 645/3507 (18.4%) OR 0.92 95% CI 0.81 to 1.04; P=0.35 Absolute difference −1.3% (95% CI −4.0 to 1.4).
All-cause hospital mortality 808/3474 (23.3%) 878/3507 (25.0%) OR 0.91 95% CI 0.81 to 1.02; P=0.27 Absolute difference −1.8% (95% CI −4.8 to 1.2).
Days alive and free of ICU (to day 90) Mean 59.4 (SD 33.9) Mean 57.8 (SD 34.2) MD 1.5 days 95% CI −0.1 to 3.0; P=0.18 Tertiary outcome; P values adjusted by Holm-Bonferroni.
Days alive and free of hospital (to day 90) Mean 42.4 (SD 32.2) Mean 40.6 (SD 32.5) MD 1.8 days 95% CI 0.3 to 3.3; P=0.08 Tertiary outcome; confidence interval excludes 0 despite adjusted P value (multiplicity control).
Days alive and free of mechanical ventilation (to day 90) Mean 64.2 (SD 34.5) Mean 62.9 (SD 34.8) MD 1.2 days 95% CI −0.4 to 2.8; P=0.28 Tertiary outcome; P values adjusted by Holm-Bonferroni.
Days alive and free of kidney replacement therapy (to day 90) Mean 71.2 (SD 32.8) Mean 70.2 (SD 33.1) MD 0.9 days 95% CI −0.6 to 2.4; P=0.25 Tertiary outcome; P values adjusted by Holm-Bonferroni.
Adverse events (any) 10/3466 (0.3%) 6/3484 (0.2%) Not reported Not reported One serious adverse event occurred in the continuous infusion group (assessed as possibly related).4
  • Primary (unadjusted) mortality analysis did not meet conventional statistical significance (OR 0.91; P=0.08), but point estimates favoured continuous infusion and the prespecified adjusted model crossed the nominal threshold (OR 0.89; P=0.04).
  • Clinical cure by day 14 was higher with continuous infusion (55.7% vs 50.0%; OR 1.26; P<0.001), whereas acquisition/colonisation/infection with multidrug-resistant organisms or C difficile was similar (7.2% vs 7.5%).
  • “Days alive and free” outcomes were directionally favourable but not statistically compelling once multiplicity adjustment was applied; adverse events were rare (<0.5%).

Internal Validity

  • Randomisation and allocation: Encrypted web-based randomisation using a minimisation algorithm stratified by study site; allocation concealment was maintained up to assignment (centralised randomisation).
  • Post-randomisation exclusions and attrition: 7202 participants were randomised; 171 (2.4%) did not consent to use of any data and were excluded from the primary (modified intention-to-treat) analyses; 50/7031 (0.7%) had missing 90-day mortality (24 continuous; 26 intermittent), yielding primary-outcome denominators 3474 and 3507.
  • Performance/detection bias: Open-label design introduces risk of differential co-interventions and biased assessment for non-mortality endpoints; this is most salient for “clinical cure” (antibiotic stopping/restarting decisions) and surveillance-dependent outcomes (e.g., multidrug-resistant organism detection).
  • Protocol adherence: Any protocol deviation occurred in 1105/3466 (31.9%) vs 1187/3484 (34.1%); incorrect assigned administration method was recorded in 282/3466 (8.1%) vs 169/3484 (4.8%); dosing/administration delays >1 hour occurred in 529/3466 (15.3%) vs 788/3484 (22.6%).4
  • Separation of the variable of interest: Prior to randomisation, most participants received intermittent infusion (77.8% vs 81.5%), with baseline use of continuous infusion already present (13.2% vs 10.1%)—a pragmatic reflection of evolving ICU practice that could dilute between-group contrast; after randomisation, “incorrect assigned administration method” deviations (8.1% vs 4.8%) indicate non-trivial contamination, likely biasing effects towards the null.4
  • Baseline characteristics: Groups were highly comparable (mean age 59.3 vs 59.6 years; mean APACHE II 19.6 vs 19.5; vasopressors/inotropes in prior 24 hours 71.0% vs 70.3%; pulmonary infection 59.0% vs 60.0%).
  • Timing and dose: Randomisation occurred after at least one β-lactam dose; time from ICU admission to randomisation had a wide distribution (median 25.2 vs 24.8 hours; IQR approximately 12–103 hours), implying heterogeneity in when “sepsis requiring β-lactam escalation” occurred during the ICU stay; dosing was clinician-determined and daily dose exposure (by defined daily dose metrics) was similar between groups on the first full day post-randomisation.
  • Heterogeneity and subgroup effects: No statistically persuasive effect modification across prespecified subgroups (pulmonary vs non-pulmonary infection, β-lactam used, age, sex, APACHE II category; interaction P values all ≥0.62), consistent with an overall modest, broadly distributed treatment effect.
  • Statistical rigour: The analysis plan was prespecified and published; multiplicity control for secondary/tertiary outcomes used Holm-Bonferroni; the primary unadjusted analysis did not achieve P<0.05 despite achieving (and exceeding) the planned sample size, emphasising the importance of predefining the estimand/modeling strategy and interpretive hierarchy.3

Conclusion on Internal Validity: Overall, internal validity is moderate-to-strong: central randomisation and an objective primary endpoint support credibility, but the open-label design, consent-related exclusions (modified intention-to-treat), and measurable protocol contamination (incorrect method use and dosing delays) plausibly attenuated between-group separation and complicate interpretation of more subjective secondary endpoints.

External Validity

  • Population representativeness: Adult ICU sepsis patients receiving two of the most common broad-spectrum β-lactams (piperacillin-tazobactam or meropenem) with substantial organ dysfunction and high use of vasopressors/respiratory support—highly aligned with “real-world” ICU sepsis case-mix in high-income settings.
  • Key exclusions limiting generalisability: Patients already requiring kidney replacement therapy at randomisation, pregnant patients, and those with limitations of life-sustaining therapy were excluded; paediatric populations were not studied.
  • Health-system applicability: Conducted across seven countries and 104 ICUs, supporting broad applicability where ICU infrastructure is similar; however, continuous infusion requires infusion pumps, compatible IV access, pharmacy/nursing workflows, and attention to drug stability (particularly for meropenem), which may constrain uptake in resource-limited environments.
  • Antibiotic ecology: The net effect of infusion strategy may vary with local pathogen MIC distributions, antimicrobial resistance prevalence, and availability of therapeutic drug monitoring; the trial was not designed to tailor dosing to pharmacokinetic targets.

Conclusion on External Validity: External validity is good for adult ICU sepsis care in systems using piperacillin-tazobactam/meropenem routinely; applicability is more limited where continuous infusion logistics or drug stability constraints are prohibitive, or where patient groups excluded here (e.g., on dialysis at baseline) dominate the target population.

Strengths & Limitations

  • Strengths:
    • Largest randomised trial to date addressing β-lactam infusion strategy in ICU sepsis, with international participation and a hard, patient-centred primary endpoint.
    • Pragmatic dosing and antibiotic choice (two common agents) maximised real-world relevance while maintaining a clean contrast on administration strategy.
    • Prespecified protocol and statistical analysis plan, including multiplicity control for non-primary outcomes.13
    • High completeness of follow-up for mortality (missing 0.7%).
  • Limitations:
    • Open-label design with potential bias for clinician-influenced endpoints (notably “clinical cure”) and for surveillance-dependent outcomes.
    • Modified intention-to-treat with exclusion of non-consenting participants post-randomisation (a small but non-zero threat to randomisation integrity).
    • Protocol deviations and contamination (incorrect assigned administration method 8.1% vs 4.8%; dosing/administration delays) could dilute the true biological effect of continuous infusion.4
    • Comparator was intermittent infusion over 30 minutes; many ICUs increasingly use extended/prolonged intermittent infusions, so BLING III does not fully answer “continuous infusion vs modern extended infusion”.

Interpretation & Why It Matters

  • Mortality signal: modest and model-sensitive
    Continuous infusion produced a small absolute mortality reduction (−1.9%) with an unadjusted primary OR of 0.91 (P=0.08); the prespecified adjusted analysis yielded OR 0.89 (P=0.04), supporting a plausible but modest benefit whose inferential strength depends on the interpretive primacy of the unadjusted vs adjusted estimand.
  • Clinical cure: improved, but potentially bias-prone
    Clinical cure increased by 5.7% (55.7% vs 50.0%); while compatible with better early infection control, it is also susceptible to open-label decision-making regarding stopping/restarting antibiotics.
  • Stewardship and safety
    No meaningful difference in new acquisition/colonisation/infection with multidrug-resistant organisms or C difficile by day 14; adverse events were uncommon (<0.5%), providing reassurance that continuous infusion is not intrinsically “toxic” when implemented with attention to dosing and stability constraints.
  • Practice implications
    Where logistics permit, BLING III supports considering continuous infusion for piperacillin-tazobactam/meropenem in ICU sepsis, particularly for patients at risk of underexposure (e.g., high renal clearance) or where MICs approach breakpoints; however, the magnitude of any mortality benefit is likely small and may be attenuated by contamination and modern comparator strategies (extended intermittent infusion).

Controversies & Subsequent Evidence

  • Primary vs adjusted inference: BLING III’s prespecified adjusted mortality analysis crossed P<0.05 (OR 0.89), whereas the unadjusted primary analysis did not (OR 0.91; P=0.08); this amplifies debate about interpretive hierarchy when adjusted models are prespecified but the canonical primary test is not statistically significant.35
  • Open-label design and “clinical cure”: The large clinical cure signal (OR 1.26) coexists with the potential for bias via clinician behaviours (e.g., timing of de-escalation, cessation, or restart), and should be weighed against the more conservative behaviour of objective endpoints (mortality, ICU/hospital-free days).
  • Contamination and workflow realities: Non-trivial protocol contamination (incorrect assigned method 8.1% vs 4.8%) and dosing/administration delays reflect real ICU operational constraints and likely bias effect estimates towards the null—raising the possibility that “true” biological benefit could be somewhat larger than observed in an explanatory trial, but also that implementation fidelity is a limiting factor.45
  • Comparator relevance in 2024+ practice: Many ICUs increasingly use extended intermittent infusions; BLING III tests continuous infusion against largely 30-minute intermittent infusion, leaving residual uncertainty about marginal gains over “extended infusion” protocols.
  • Major follow-up RCT evidence: The MERCY trial (continuous vs intermittent meropenem in sepsis/septic shock) reported no statistically significant reduction in 28-day mortality, but effect estimates were directionally concordant with a small benefit—contextualising BLING III as part of an accumulating “modest benefit” signal rather than a single decisive result.65
  • Systematic review/meta-analysis synthesis: A contemporary Bayesian meta-analysis including BLING III (which contributed most statistical weight) estimated reduced mortality with prolonged infusion (RR 0.86; 95% credible interval 0.72 to 0.98; high certainty) and improved clinical cure (RR 1.16; 95% credible interval 1.07 to 1.31), with no clear increase in adverse events—supporting the view that the “true” effect is likely small but real.7
  • Guideline landscape: The Surviving Sepsis Campaign 2021 guidelines suggested prolonged infusion for β-lactams in sepsis/septic shock (weak/conditional), reflecting pre-BLING III uncertainty; BLING III and subsequent syntheses are likely to strengthen evidentiary confidence in future updates.25

Summary

  • BLING III randomised 7202 ICU patients with sepsis across 104 ICUs to continuous vs intermittent infusion of piperacillin-tazobactam or meropenem.
  • The unadjusted primary analysis showed no statistically significant mortality reduction at day 90 (24.9% vs 26.8%; OR 0.91; P=0.08), though the prespecified adjusted analysis favoured continuous infusion (OR 0.89; P=0.04).
  • Clinical cure by day 14 improved with continuous infusion (55.7% vs 50.0%; OR 1.26), without a detectable increase in multidrug-resistant organism/C difficile acquisition (7.2% vs 7.5%).
  • Protocol contamination and open-label design likely attenuated separation and increase susceptibility to bias for non-mortality endpoints.
  • Subsequent evidence synthesis suggests the overall effect of prolonged/continuous infusion is probably modest but beneficial on average, with no clear safety penalty.

Further Reading

Other Trials

Systematic Review & Meta Analysis

Observational Studies

Guidelines

Notes

  • Secondary and tertiary outcome P values were adjusted using a Holm-Bonferroni procedure, whereas confidence intervals in the main report are presented without multiplicity adjustment (a common but interpretively important asymmetry).

Overall Takeaway

BLING III is the largest and most methodologically robust RCT to evaluate β-lactam infusion strategy in ICU sepsis, demonstrating a consistent but modest signal towards improved outcomes with continuous infusion—most clearly for clinical cure, and more ambiguously for mortality (borderline in the unadjusted primary analysis, nominally significant in the prespecified adjusted analysis). In conjunction with contemporary synthesis evidence, it shifts the evidentiary centre of gravity towards prolonged/continuous infusion as a reasonable default where operationally feasible, while underscoring that any survival benefit is likely small and implementation fidelity is crucial.

Overall Summary

  • Continuous β-lactam infusion in ICU sepsis produced a small, directionally favourable mortality effect (primary P=0.08; prespecified adjusted P=0.04) and improved clinical cure, with rare adverse events and no clear increase in resistance/C difficile acquisition.

Bibliography