A systematic review and meta-analysis of randomized controlled trials comparing short-course antibiotic therapy, defined as less than 14 days, with standard-course therapy of 14 days or more for neonatal sepsis has concluded that the available evidence is insufficient to determine whether the shorter strategy is non-inferior. The review, published in Frontiers in Pediatrics on 28 October 2026, pooled data from just three randomized controlled trials covering 434 neonates, all conducted in India. The pooled risk ratio for the primary composite outcome of treatment failure or mortality was 0.995, with a 95 percent confidence interval stretching from 0.144 to 6.890 and a p-value of 0.993. Although that point estimate sits almost exactly at no difference, the authors emphasize that the extreme width of the interval means the result cannot reliably confirm equivalence or rule out substantial benefit or harm.

Neonatal sepsis remains a significant cause of morbidity and mortality worldwide, with reported incidence rates as high as 4.3 to 6.5 per 1,000 live births and a high prevalence of low birth weight and premature infants among affected cases. The etiological landscape spans early-onset and late-onset disease, with pathogens ranging from Gram-positive organisms such as Streptococcus agalactiae and coagulase-negative staphylococci to Gram-negative bacilli including Escherichia coli, Klebsiella pneumoniae and Pseudomonas aeruginosa. The emergence of multidrug-resistant organisms, including metallo-beta-lactamase-producing P. aeruginosa, adds urgency to the question of how long antibiotics should be given. Current treatment durations are often prolonged, typically 14 days or more, based on historical practice rather than robust randomized evidence, and extended exposure may increase the risk of antibiotic-associated adverse events, promote resistance and prolong hospital stays.

The principle of using the shortest effective antibiotic course is increasingly advocated across infectious diseases to mitigate antimicrobial resistance, disruption of the microbiome and increased healthcare costs. Evidence from other clinical contexts, including community-acquired pneumonia and bacteremia, suggests shorter courses can be as effective as longer regimens without compromising safety. The research team, from the Guangdong-Hongkong-Macau Institute of CNS Regeneration at Jinan University and the Department of Pediatrics at Zhujiang Hospital of Southern Medical University in Guangzhou, China, sought to apply the same rigor to neonates. They searched Web of Science, PubMed, EMBASE and the Cochrane Central Register of Controlled Trials on April 19, 2026, including only randomized controlled trials with a minimum follow-up of 28 days after antibiotic completion or until hospital discharge, and assessed study quality with the Cochrane Risk of Bias 2 tool.

The three included trials differed in design and population. Jain, 2026, conducted in central India, enrolled 140 neonates with uncomplicated, culture-proven sepsis weighing at least 1,000 grams at birth who were clinically improving after an initial seven days of sensitive antibiotic therapy, randomizing 70 to a total of seven days of antibiotics and 70 to 14 days; it was an assessor-blinded non-inferiority trial whose primary outcome was a composite of probable or definitive relapse of sepsis within 21 days of antibiotic completion. Islam, 2023, a two-arm randomized non-inferiority trial in a level III NICU in West Bengal and surrounding districts, randomized 234 newborns who were clinically improving at day nine of intravenous antibiotic therapy, with the study group stopping at a total of 10 days and the control group continuing to 14 days; allocation used a computer-generated sequence with stratification by birth weight and gestational age, concealed in serially numbered, sealed opaque envelopes. Gathwala, 2010, included 60 clinically symptomatic neonates with blood culture-proven sepsis, comparing a 10-day course with a conventional 14-day course.

Across the pooled trials, statistical heterogeneity was absent, with an I2 of 0 percent, a tau2 of 0 and an H statistic of 1, indicating that variability among the three point estimates was consistent with sampling error alone and that a fixed-effect model was appropriate. But the authors stress that this homogeneity does not confer precision. The severe imprecision of the individual studies, a consequence of small sample sizes, propagates directly into the pooled estimate. The confidence interval for the primary outcome encompasses both a substantial relative risk reduction of roughly 86 percent and a nearly seven-fold increase in risk, a range so broad that any clinical inference is impossible. The test statistic was Z = −0.010, corroborating the null point estimate while underscoring that the analysis cannot distinguish between equivalence, benefit or harm.

A post-hoc exploratory leave-one-out sensitivity analysis exposed how fragile the pooled result is. Omitting Islam, 2023, shifted the summary estimate to 0.60 (95 percent CI 0.012 to 30.363; p = 0.349). Omitting Jain, 2026, shifted it in the opposite direction to 1.653 (95 percent CI 0.032 to 84.324; p = 0.351). Omitting Gathwala, 2010, left the estimate virtually unchanged at 0.996 but produced a confidence interval spanning from 0.00015 to 6,654, with a small amount of between-study variance emerging (tau2 = 0.234, I2 = 0.243 percent). The authors note that the exclusion of any single study caused dramatic and erratic shifts in both the point estimate and the already extreme confidence intervals, meaning the summary statistic is entirely dependent on the specific combination of these few small trials and that a single new, similarly sized trial could drastically alter the conclusion.

Risk of bias assessment added further caution. Only one study, Islam, 2023, was judged to have a low overall risk of bias. Jain, 2026, was assessed as having some concerns overall, including in the domains of deviations from intended interventions and selection of the reported result, while Gathwala, 2010, was judged to have an unknown overall risk, primarily because insufficient information in a key domain prevented a definitive judgment, particularly regarding whether outcome assessors were blinded. Randomization and handling of missing outcome data were rated low risk across all three studies, but the authors point out that in an outcome like treatment failure or mortality, which can involve subjective clinical judgments, lack of blinding can systematically influence results, potentially masking a true difference between short and standard courses or creating a spurious one.

Publication bias could not be reliably assessed. Egger’s linear regression test yielded an intercept of −0.061918 with a standard error of 3.1033, a t-value of 0.018869 and a non-significant p-value of 0.988, while Begg’s rank correlation test produced a Kendall’s score of 1, a Z-value of 0.52223 and a p-value of 0.6015. With only three studies, the funnel plot lacks the scatter needed to form a recognizable shape, and the authors state that the power of these formal tests under such conditions is virtually nil. The Trim and Fill method, however, imputed one missing study and adjusted the pooled effect from 0.995 to 0.5, with a 95 percent confidence interval of 0.11764 to 2.1252, suggesting possible asymmetry while remaining statistically non-significant. The authors treat this as an exploratory signal that the published literature may over-represent null findings, not as evidence of a true benefit.

The review’s conclusion is deliberately restrained: current evidence is insufficient to determine the non-inferiority of short-course therapy due to very low precision, a limited number of studies and methodological concerns. The authors frame the finding not as an inconclusive result but as a defined research gap, arguing that the apparent equipoise reflected in the point estimate is in reality an equipoise of ignorance. They call for large, multicenter, pragmatic randomized controlled trials with low risk of bias and sufficient power to yield a precise and clinically actionable estimate, and suggest future work should explore optimal minimum effective durations for specific clinical and microbiological subgroups, investigate long-term ecological impacts on resistance patterns and microbiome development, and integrate rapid molecular diagnostics to enable pathogen-directed duration decisions. Until such trials are completed, the question of optimal antibiotic duration for neonates with sepsis remains unanswered by existing randomized evidence.

Subject of Research: Comparison of short-course versus standard-course antibiotic therapy for neonatal sepsis in a systematic review and meta-analysis of randomized controlled trials

Article Title: Efficacy and safety of short-course (<14 days) vs. standard-course (≥14 days) antibiotic therapy for neonatal sepsis: a systematic review and meta-analysis

Article References: Siqi, L., & Zeng, Q. (2026). Efficacy and safety of short-course (<14 days) vs. standard-course (≥14 days) antibiotic therapy for neonatal sepsis: a systematic review and meta-analysis. Frontiers in Pediatrics, 14, Article 1956188. https://doi.org/10.3389/fped.2026.1956188

Image Credits: AI Generated

DOI: 10.3389/fped.2026.1956188

Keywords: neonatal sepsis, antibiotic duration, meta-analysis, randomized controlled trials, antimicrobial stewardship, treatment failure, risk of bias, publication bias, systematic review, infant mortality, antimicrobial resistance, Efficacy