Among 3,669 ICU Patients With Sepsis, Beta-Lactam Antibiotics Were Associated With 12% Lower In-Hospital Mortality Than Glycopeptides. Whether the First Dose Arrived at Hour 2 or Hour 47 Made No Measurable Difference.

Sepsis accounts for roughly 20% of all deaths worldwide, and for over a decade the dominant clinical mandate has been speed: get antibiotics into a septic patient as fast as possible, ideally within one hour of recognition. That mandate has shaped hospital sepsis bundles, quality metrics, and in many systems, financial penalties tied to door-to-antibiotic time. What has received comparatively little rigorous causal attention is a different and arguably more clinically actionable question: among the antibiotics a physician could reasonably choose empirically, does the specific class matter more than the minutes on the clock? This study, using the MIMIC-IV critical care database and a modern causal inference technique called target trial emulation, asks exactly that question, and its answer complicates two decades of sepsis quality improvement orthodoxy.

Li and colleagues used the MIMIC-IV database, a large, de-identified, publicly available critical care dataset, to emulate a target trial, a method that applies the rigor of a randomized controlled trial's design principles to observational data specifically to correct for confounding by indication. Confounding by indication is the core problem with most retrospective antibiotic timing studies: sicker patients tend to get antibiotics faster because clinicians recognize how sick they are, which can make faster treatment look artificially harmful or obscure a true benefit. To address this, the researchers used a clone-censor-weight approach, a technique that creates cloned copies of each patient following different hypothetical treatment strategies, then statistically weights and censors those clones to approximate what a true randomized trial would have shown. The study population consisted of ICU patients with sepsis who received either a beta-lactam or a glycopeptide (such as vancomycin) as their first-line empirical antibiotic. The researchers compared antibiotic initiation within 48 hours of sepsis diagnosis against delayed initiation, and separately compared beta-lactam against glycopeptide therapy, using weighted Cox regression to estimate hazard ratios for in-hospital mortality, with sensitivity analyses stratified by age, sex, and two standard ICU severity scores, the SOFA and SAPS II scores.

Among 3,669 eligible ICU patients with sepsis, 3,568 (97%) received antibiotics within 48 hours of diagnosis. After weighting to achieve covariate balance between comparison groups, beta-lactam use was associated with significantly lower in-hospital mortality than glycopeptide use (hazard ratio 0.88, 95% CI 0.78 to 0.95), a 12% relative reduction. That mortality benefit remained consistent when the researchers looked at 7-day, 14-day, and 60-day mortality separately. In contrast, timing of antibiotic initiation within the 48-hour window showed no significant association with outcomes for either beta-lactams or glycopeptides. Sensitivity analyses stratifying by age, sex, SOFA score, and SAPS II score produced wider confidence intervals in some subgroups but did not change the direction or overall magnitude of either finding.

• —This is a retrospective, observational analysis using electronic health record data from a single health system's ICU database, MIMIC-IV. Despite the rigor of target trial emulation, it cannot fully replace a genuine randomized controlled trial directly comparing empirical beta-lactam versus glycopeptide strategies in sepsis.
• —The comparison groups were determined by clinician choice of empirical antibiotic, and while the clone-censor-weight method is specifically designed to correct for confounding by indication, some unmeasured confounding, such as suspected pathogen or infection source not fully captured in the dataset, cannot be entirely excluded.
• —With 97% of the cohort receiving antibiotics within 48 hours, statistical power to detect a true timing effect outside that narrow window, particularly for delays beyond 48 hours, is limited by how few patients experienced meaningfully delayed treatment.
• —The study compared beta-lactams broadly against glycopeptides broadly. It does not resolve which specific beta-lactam, dose, or infusion strategy drives the observed benefit, a distinction that matters considerably for translating this finding into a specific prescribing protocol.

Antibiotic class choice, not the number of minutes before the first dose, was the variable associated with a real difference in sepsis mortality in this analysis, once a reasonably prompt standard of care was already being met. That does not license complacency about delaying antibiotics indefinitely, but it does suggest that the intense institutional focus on shaving minutes off door-to-antibiotic time may have outpaced the evidence for what that speed alone actually buys a patient, while the choice between a beta-lactam and a glycopeptide, a decision made in the same clinical moment, may carry more weight than the clock ever did.