Efficacy of Antibiotic Regimens for Sepsis or Possible Serious Bacterial Infection in Young Infants Aged 0 to 59 Days: A Systematic Review and Meta-analysis

The protocol for this systematic review was prospectively registered on PROSPERO (CRD42023417564).¹⁸ Our population, intervention, comparator, outcome (PICO) question was: Among young infants aged 0 to 59 days with sepsis or PSBI, (P) do antibiotic regimens for the treatment of sepsis or PSBI, (I) compared with alternate antibiotic regimens or alternate management such as observation, referral, or placebo, (C) improve critical outcomes (O)?

This systematic review was designed as an update to the previous Cochrane reviews of neonatal sepsis. It was commissioned by the WHO to inform guidelines regarding treatment of young infant sepsis.

We systematically searched MEDLINE, Embase, CINAHL, the WHO Global Index Medicus, and the Cochrane Central Registry of Trials. The search included terms for infant, sepsis, integrated management of childhood illness infection signs, antibiotics, and clinical trials. The full search strategy is available in Supplemental Appendix 1. This systematic review was designed as an update to 3 prior Cochrane systematic reviews of the treatment of EOS, LOS, and community-based sepsis in neonates^15–17; as such, we searched the published literature since the last prior Cochrane search date from January 1, 2018 to April 18, 2023 to encompass dates starting from the earliest search date of the previous reviews. We included all studies that were included in the previous Cochrane reviews. We also hand-searched the bibliographies and lists of excluded studies from the Cochrane reviews and other key systematic reviews related to young infant sepsis to identify additional studies that met our inclusion criteria.

We restricted inclusion to published randomized controlled trials (RCTs). We excluded unpublished gray literature and abstracts for which there was no full text. We applied no limitations based on language or publication date.

We did not limit inclusion based on the study setting. Enrollment and treatment could occur in any location, including the home, health center, outpatient health facility, or hospital. We defined nonhospital-based trials as those in which infants received their antibiotic treatment in any setting not self-identified as a “hospital” facility, including a health center, health post, or the home. We defined hospital-based trials as those in which infants received their antibiotic treatment while inpatient in a hospital, as defined by the study investigators.

We included trials of infants 0 to 59 days postnatal age who presented with suspected sepsis or PSBI. Studies enrolling infants older than 60 days were included only if: (1) there was a subgroup analysis presenting outcomes among infants aged 0 to 59 days or (2) the mean postnatal age in days plus 1 SD above the mean was under 60 days.

We did not apply an a priori definition for sepsis or PSBI or require trials to employ a specific definition. Sepsis and PSBI definitions varied by individual trial; definitions included clinical exam criteria, laboratory markers, and blood cultures. PSBI is defined by 3 subgroups of infants: critical illness, clinical severe infection, and isolated fast breathing.⁵ Clinical severe infection is defined by the WHO as at least 1 sign of severe infection (movement only when stimulated, not feeding well, temperature ≥38°C or ≤35.5°C, or severe chest indrawing) among infants aged 0 to 59 days. Isolated fast breathing is treated as clinical severe infection among infants aged 0 to 6 days who present with isolated tachypnea, defined as a respiratory rate >60 breaths per minute. Critical illness is defined as the presence of any of the following signs among infants aged 0 to 59 days: convulsions, inability to feed, or no movement on stimulation.¹⁹ We excluded studies that limited enrollment to infants who were treated prophylactically or based on risk factors with no clinical evidence of sepsis.⁵ 

We included trials evaluating any antibiotic regimen intended to treat young infant sepsis. We excluded studies of antiviral, antifungal, or antimalarial interventions. We excluded packages of general newborn care in which antibiotics were a component of several interventions, and in which the antibiotic effect could not be isolated.

We included studies with the comparator group of any antibiotic regimen or alternative treatment strategy. For example, a standard of care antibiotic regimen, an alternate antibiotic regimen, an alternate management strategy such as referral or observation, or placebo. The comparator antibiotic regimen could differ from the intervention by type of antibiotic, dose, duration, or mode of delivery.

Critical outcomes included mortality, treatment success or failure, neurodevelopmental outcomes, and adverse effects. Secondary outcomes included hospital utilization, such as length of stay or readmission, relapse, and cost-effectiveness. We preferentially reported the latest time point available for all outcomes. Studies reporting only pharmacokinetic outcomes were excluded.

We initially screened studies based on title and abstract, then by full text. All study screening was conducted in duplicate and adjudicated by a third author using Covidence systematic review software.²⁰ 

We extracted data into a predesigned data extraction form. All data were independently extracted in duplicate with discrepancies adjudicated by a third author. For dichotomous outcomes, we extracted event rates (n/N, %). For continuous outcomes, we extracted mean per SD or median per interquartile range. We also extracted effect sizes (unadjusted and adjusted), intention-to-treat (ITT) and per-protocol (PP) outcomes.

Risk of bias was assessed using the revised Cochrane risk-of-bias tool for randomized trials (RoB 2) for individually randomized trials,²¹ and the revised Cochrane risk-of-bias tool for cluster-randomized trials (RoB 2 CRT).²² Risk of bias was assessed by 2 independent assessors; discrepancies were resolved by consensus. A study was deemed to have an overall high risk of bias if there was high risk in any 1 domain or some risk in 3 or more domains. A study was deemed to have overall some risk of bias if there was some risk in 1 or 2 domains. A study was deemed to have overall low risk of bias if there was low risk in every domain. In the case of more than 10 studies of a common intervention reporting a common outcome, we planned to assess for potential publication bias per Cochrane methods using funnel plots and Egger’s tests for dichotomous outcomes and meta-regression for continuous outcomes.

Outcome data were summarized in study data tables that were grouped by outcome. Where there were sufficiently similar interventions and outcomes to perform pooled analysis for outcomes of interest, we planned to conduct inverse-variance, random-effects meta-analysis to generate forest plots and a pooled effect size.²³ Two studies were the minimum number required to conduct meta-analysis and generate forest plots.

We calculated both relative and absolute effect sizes for critical outcomes. For dichotomous outcomes, we calculated the relative risk (RR) per 95% confidence interval (CI) and risk difference (RD) per 95% CI. For continuous outcomes, we calculated the mean difference (MD) per 95% CI. We preferentially used ITT outcomes over PP outcomes for calculating effect sizes and determining certainty of evidence when data were available for both.

Certainty of evidence (COE) for the effects of interventions on individual outcomes was assigned using Grading of Recommendations Assessment, Development, and Evaluation methods²⁴ and categorized as very low, low, moderate, or high COE. For each outcome, we determined the COE based on risk of bias, inconsistency, indirectness, imprecision, and presence of publication bias. See Supplemental Appendix 2 for details. We created Summary of Findings tables with COE for WHO recommended regimens (first and second choice) compared with alternate regimens. Antibiotic duration regimens are provided in the appendix. We did not determine the COE and only report effect sizes as reported by the trials for comparisons of antibiotic regimens that included antibiotics or medications not listed on the WHO Essential Medicines List for Children.²⁵ 

Data were sufficient to conduct subgroup analyses based on World Bank country income level²⁶ and hospital versus nonhospital settings. We did not have sufficient data to conduct subgroup analyses on the other a priori subgroups of interest: phase of illness (at presentation versus after 48–72 hours of treatment), definition of sepsis (eg, blood culture confirmed), EOS versus LOS, preterm versus full-term gestational age, timing of antibiotic administration, antimicrobial resistance setting, and studies with packages or care that included antibiotics versus antibiotics alone. A minimum of 3 studies were required to conduct subgroup analyses.²³ 

All databases were searched from January 1, 2018 to April 18, 2023. After deduplication, we screened the 2881 study titles and abstracts from our database searches and 109 studies from snowballing other systematic reviews including trials published before 2018. We reviewed the full texts of 96 studies (Fig 1). Forty-one trials met criteria for inclusion; 31 of these trials were not included in the previous Cochrane reviews. The most common reason for exclusion at the full text level was study design, in that the study was not an RCT (n = 35) (see Supplemental Table 4 for excluded studies).

Study publication years ranged from 1978 to 2022. Among the 41 included trials, 1 was a cluster-randomized crossover trial and 40 were individually randomized. Thirty-five trials were hospital-based and 6 were nonhospital based. The 41 trials comprised a total of 18 054 participants. All studies enrolled infants at clinical presentation; no studies addressed antibiotic regimens for treatment failure after 48 to 72 hours of initial antibiotic therapy, or time to first antibiotic dose administration after sepsis identification.

Twenty-one studies had overall high risk of bias, 18 studies had overall some risk of bias, and 2 studies had overall low risk of bias (Fig 2: summary graph, Supplemental Table 11: detailed risk of bias assessment by domain and study). For the domain of “bias arising from the randomization process”, 2 trials were deemed high risk because of the use of predictable, unblinded methods of randomization such as alternating assignment to treatment groups. Eleven trials were deemed to be at some risk of bias, primarily because of lack of information about allocation concealment during randomization. For the domain of “bias because of deviations from the intended interventions,” 4 studies were deemed to be at high risk. Twenty studies were deemed to have some risk, primarily because of lack of information about protocol deviations with no clear study flow diagram. For the domain of “bias because of missing outcome data,” 5 studies were deemed to be at high risk and 10 studies were deemed to be at some risk of bias. Most studies were low risk for this domain, likely because of the predominance of short-term primary outcomes that occurred during the initial hospitalization. For the domain of “bias in measurement of the outcome,” 12 studies were deemed to be at some or high risk of bias, largely because of ill-defined or subjective primary outcomes with unblinded outcome assessors. For the domain of “bias in selection of the reported result,” 1 study was deemed to be at high risk and 25 were deemed to be at some risk, primarily because of a lack of information related to trial registration, no protocol paper, and no clear reference to a predetermined data analysis plan. Overall, most trials assigned a composite high risk of bias were lacking critical information in multiple domains such as protocol deviations, missing outcomes, or a prespecified data analysis plan.

Detailed individual-level trial characteristics of the 35 hospital-based trials are shown in Table 2. Seventeen trials assessed WHO-recommended regimens (first- and second-choice) compared with alternate regimens, 6 assessed other non-WHO-recommended regimens, and 12 trials assessed antibiotic duration or dosing frequency.

Infant age at enrollment varied, but 3 trials only enrolled infants with EOS (Metsvaht 2010; Snelling 1983; Tewari 2014),^27–29 and 4 trials only enrolled infants with LOS (Ceriani Cernadas 2014; Lutsar 2020; Ramasamy 2014; Shabaan 2017).^30–33 Only 1 trial limited enrollment to preterm infants (Millar 1992),³⁴ though 1 trial reported a subgroup analysis of extremely low-birth weight infants (Metsvaht 2011).³⁵ Thirty-one trials were conducted in level III or tertiary centers.

The definition of sepsis varied widely among the hospital-based trials. Twenty trials incorporated a combination of clinical signs or symptoms, perinatal risk factors, and laboratory markers to diagnose suspected sepsis (Table 2). Six trials required a positive blood culture for enrollment.^33,36–40 Among an additional 27 trials conducting blood cultures as part of routine care, the rate of culture positivity among infants with suspected sepsis varied from 2.7%²⁹ to 65.7%.³⁶ The pathogens most commonly reported across trials were Staphylococcus aureus and Escherichia coli (Supplemental Table 12).

Eleven trials compared the WHO-recommended first-choice regimen for hospital treatment of young infant sepsis (IV or IM ampicillin or penicillin plus gentamicin) with any alternate regimen.^{28,31,35–37,41–46,}Tables 3 and Supplemental Table 13 contain a summary of findings and Supplemental Tables 5–10 contain detailed, study-level outcomes. Five trials compared the WHO second-choice regimen of a third-generation cephalosporin to the WHO first-choice regimen of ampicillin or penicillin plus an aminoglycoside, typically gentamicin (summary of findings shown in Table 3; forest plots shown in Supplemental Fig 3).^{28,36,41–43} There were no differences between cephalosporins versus ampicillin or penicillin plus an aminoglycoside in likelihood of treatment failure (risk ratio [RR] 1.02 [95% CI 0.38 to 2.71], 2 trials, n = 955; very low COE), treatment success (RR 1.03 [95% CI 0.93 to 1.13], 4 trials, n = 1083; moderate COE), mortality (RR 0.64 [95% CI 0.25 to 1.65], 3 trials, n = 711, very low COE), or neurologic sequelae (RR 1.39 [95% CI 0.69 to 2.78], 1 trial, n = 170, very low COE) (Table 3, Supplemental Fig 3). Cephalosporins were associated with lower likelihood of any adverse events, including jaundice (RR 0.35 [95% CI 0.15 to 0.82], 3 trials, n = 628, moderate COE). The subgroup analysis by income setting similarly showed no differences in effect on treatment success between high-income countries (RR 1.03 [95% CI 0.98 to 1.08], 2 trials, n = 720) and LMIC settings (RR 1.19 [95% CI 0.70 to 2.03], 3 trials, n = 363).

Six additional trials compared alternate regimens to ampicillin or penicillin plus gentamicin.^{27,31,37,45–47} One cluster-randomized trial of 283 infants in Estonia (Metsvaht 2010) compared 2 WHO first-choice regimens to each other—ampicillin plus gentamicin versus penicillin plus gentamicin—and concluded that the regimens had similar likelihood of treatment failure (adjusted odds ratio [aOR] 1.01 [95% CI 0.52 to 1.97], clusters = 2, very low COE) and mortality (aOR 0.76 [95% CI 0.33 to 1.75], clusters = 2, very low COE).²⁷ A subgroup analysis of 75 extremely low birth weight infants (birth weight <1000 g) enrolled in this trial (Metsvaht 2011) found a similar lack of difference between study arms for both treatment failure (aOR 1.1 [95% CI 0.4 to 3.1]) and mortality (aOR 0.90 [95% CI 0.3 to 2.9]) (Supplemental Tables 5 and 6, 13).³⁵ This study did not report on the critical outcomes of treatment success, neurologic sequelae, or adverse events. One individually-randomized trial (n = 120) compared standard of care (typically ampicillin plus gentamicin) plus fosfomycin versus standard of care alone in a rural Kenyan hospital with growing antimicrobial resistance (Obiero 2022).⁴⁴ Mortality (RR 1.29 [95% CI 0.30 to 5.52], n = 120, very low COE), adverse events (RR 0.97 [95% CI 0.47 to 2.00], n = 107, very low COE), and readmission rates (RR 0.14 [95% CI 0.01 to 2.62], n = 120, very low COE) were similar between study groups, though estimates were imprecise. This study did not report on the critical outcomes of treatment success, treatment failure, or neurologic sequelae. Another recent individually-randomized trial (n = 271) compared meropenem to standard care (ampicillin or cefotaxime plus gentamicin) for LOS in 18 European NICUs with high rates of documented antimicrobial resistance (Lutsar 2020).³¹ Only 32 of 63 (51%) positive cultures in the meropenem arm and 32 of 77 (42%) in the standard care arms were susceptible to the allocated antibiotics. Mortality (RR 1.42 [95% CI 0.56 to 3.62], n = 271, very low COE), relapse (RR 0.89 [95% COE 0.32 to 2.46], n = 75, very low COE), treatment success (RR 1.41 [95% CI 0.95 to 2.09], n = 271, very low COE), and adverse events (RR 0.87 [95% CI 0.75 to 1.01], n = 268, very low COE) were similar between groups.³¹ This study did not report on the critical outcomes of treatment failure or neurologic sequelae. Three additional trials compared ampicillin and gentamicin to other alternative regimens (azlocillin plus gentamicin, mezlocillin, cephalothin plus tobramycin, and ampicillin plus amikacin) but these studies were all over 35 years old at the time of this review and reported on antibiotics not in the WHO Essential Medicines List.^37,45,46 

Seven hospital-based trials studied the efficacy of WHO-recommended second-choice antibiotic regimens (third-generation cephalosporins, piperacillin-tazobactam, oxacillin plus aminoglycoside, and vancomycin plus aminoglycoside).^{29,30,32,34,48–50} Trials were not able to be pooled and individual level results are shown in Tables 2 and 3 and Supplemental Tables 5–10 and 13. One trial (Tewari 2014) compared amikacin to piperacillin-tazobactam in a NICU in India with high rates of resistance to ampicillin, gentamicin, and cefotaxime.²⁹ This trial found no differences between amikacin versus piperacillin-tazobactam in risk of mortality (RR 3.03 [95% CI 0.13 to 73.48], n = 187, very low COE), treatment failure (RR 1.52 [95% CI 0.26 to 8.87], n = 187, very low COE), relapse (RR 1.01 [95% CI 0.15 to 7.03], n = 187, very low COE), or adverse events (RR 0.34 [95% CI 0.01 to 8.16], n = 187, very low COE), and did not report on the critical outcomes of treatment success or neurologic sequelae. Another trial (Ramasamy 2014) compared cefotaxime plus gentamicin versus cloxacillin plus amikacin for LOS in India, finding no differences in risk of mortality (RR 2.67 [95% CI 0.79 to 9.04], n = 90, very low COE), treatment failure (RR 3.20 [95% CI 0.37 to 27.51], n = 90, very low COE), hospitalization (RR 0.50 [95% CI 0.09 to 3.04], n = 90, very low COE), neurologic sequelae (RR 4.50 [95% CI 1.04 to 19.39], n = 72, low COE), or adverse events (RR 4.00 [95% CI 0.49 to 32.87], n = 187, very low COE).³² This trial did not report on treatment success. One trial in an Argentine NICU with high rates of coagulase-negative Staphylococcus (Ceriani Cernadas 2014) found similar rates of treatment success comparing cefazolin plus amikacin versus vancomycin plus amikacin (RR 1.07 [95% CI 0.94 to 1.22], n = 109, moderate COE), no difference in risk of adverse events (RR 0.36 [95% CI 0.02 to 8.76], n = 109, very low COE), and did not report on mortality, treatment failure, or neurologic sequelae.^30,34,51 Four studies reported on antibiotics not included in the WHO Essential Medicines List.^34,48–50 

Six trials compared non-WHO recommended regimens versus other non-WHO recommended regimens (Table 2, Supplemental Tables 5–10).^45,51–55 All of these comparisons included antibiotics or medicines not included in the WHO Essential Medicines List. Overall, comparisons were heterogeneous and insufficient to support conclusions regarding the efficacy of one regimen over another.^45,51–55 

Seven trials compared different durations of antibiotic therapy for the treatment of sepsis (Tables 2 and 3, Supplemental Tables 5–9, and 14).^{38–40,56–59} Most of these studies were relatively small (range: n = 60 to 504; median: n = 116). Five trials in India limited randomization to infants with culture-proven sepsis for whom symptoms resolved within 48 hours of antibiotic initiation. One trial in Pakistan (Anwar 2020) enrolled infants with an elevated C-reactive protein that was indicative of possible sepsis. When comparing 3 versus 7 days of antibiotic treatment, Anwar (2020) found a similar likelihood of treatment failure between groups (RR 0.78 [95% CI 0.31 to 1.95]), n = 120, very low COE) but did not report on treatment success, mortality, adverse events, or neurologic sequelae.⁵⁷ Comparing 7 vs 10 days of treatment, Rohatgi (2017) found no difference in likelihood of treatment failure (RR 1.00 [95% CI 0.06 to 15.64], n = 128, very low COE), hospital readmission (RR 3.00 [95% CI 0.32 to 28.10], n = 128, very low COE), or hospital length of stay (MD 0.60 days [95% CI −0.61 to 1.81], n = 128 low COE), and did not report on treatment success, mortality, adverse events or neurologic sequelae.³⁸ In the comparison of 7 vs 14 days antibiotic duration, Chowdhary (2006) found similar likelihood of treatment failure between groups (RR 6.18 [95% CI 0.78 to 48.64], n = 69, very low COE), whereas RR of adverse events was not estimable because of lack of events in both groups.⁵⁸ The study did not report on treatment success, mortality, or neurologic sequelae. In a pooled analysis of 3 trials, 10 versus 14 days of treatment resulted in similar likelihood of treatment failure (RR 1.60 [95% CI 0.20 to 12.59], 3 trials, n = 356, low COE) but length of hospital stay was shorter among the infants receiving shorter duration antibiotic therapy (MD −5.22 days [95% CI −7.4 to −3.0], 2 trials, n = 130, high COE) (Reddy 2022; Gathwala 2010; Fursule 2022; Supplemental Fig 4).^39,40,59 None of the 3 trials comparing 10 versus 14 days of treatment reported on outcomes of treatment success, adverse events, or neurologic sequelae, and pooled likelihood of mortality was not estimable. Another study (Keij 2022) compared 7 days of intravenous (IV) therapy versus 2 to 3 days of IV plus 4 to 5 days of PO therapy among hospitalized infants in the Netherlands and found similar rates of serious adverse events (RR 1.16 [95% CI 0.64 to 2.09], n = 504, very low COE), treatment failure (RR 5.00 [95% CI 0.24 to 103.63], n = 504, very low COE), and hospital readmission (RR 1.11 [95% COE 0.46 to 2.69], n = 504, very low COE) between groups, and did not report on treatment success, mortality, or neurologic sequelae.⁵⁶ 

Five trials compared different antibiotic dosing frequency or duration of dose administration.^33,60–63 Individual trial characteristics and outcomes are shown in Tables 2 and 3 and Supplemental Tables 5–9 and 15. Three aminoglycoside studies tested the safety and efficacy of daily compared with twice or 3 times daily dosing.^60,62,63 Of 2 studies testing gentamicin dosing frequency, English (2004) found similar rates of nephrotoxicity (RR 0.48 [95% CI 0.04 to 5.17], 1 trial, n = 297, very low COE) and mortality (RR 1.07 [95% CI 0.59 to 1.92], 1 trial, n = 297, low COE) between dosing groups, and Kosalaraksa (2004) found similar likelihood of treatment success (RR 1.08 [95% CI 0.97 to 1.20], 1 trial, n = 51, moderate COE).^62,63 One study testing amikacin dosing frequency (Abdel-Hady 2011) found similar time to treatment success, defined as time to normalization of C-reactive protein (MD −2.30 days [95% CI −6.51 to 1.91], n = 30, very low COE) and time to normalization of complete blood count (MD 0.70 days [95% CI −2.01 to 3.41], 1 trial, n = 30, very low COE), between groups.⁶⁰ Of these 3 aminoglycoside dosing studies, only English (2004) reported on mortality or adverse events (nephrotoxicity), and none reported on the critical outcomes of treatment failure or neurologic sequelae.

One study (Gwee 2019) compared continuous vancomycin infusions to intermittent infusions and found no difference in the rate of nephrotoxicity (RR 0.94 [95% CI 0.06 to 14.63], n = 104, very low COE), whereas RRs for mortality and drug-related adverse events were not estimable because of no or negligible events in either arm.⁶¹ The study did not report on critical outcomes of treatment failure or success, or neurologic sequelae. One study (Shabaan 2017) compared prolonged meropenem infusion to intermittent infusion among infants with late-onset gram-negative sepsis.³³ The prolonged infusion group had lower rates of mortality (RR 0.44 [95% CI 0.20 to 0.97], n = 102, low COE) and adverse events (RR 0.43 [95% CI 0.25 to 0.75], n = 102, moderate COE) and higher treatment success rates, defined as both clinical improvement (RR 1.82 [95% CI 1.17 to 2.85], n = 102, low COE) and negative culture 7 days after starting treatment (RR 1.45 [95% CI 1.10 to 1.90], n = 102, moderate COE) than the intermittent infusion group, but did not report on critical outcomes of treatment failure, or neurologic sequelae.

Six nonhospital-based studies compared the effects of WHO first-choice antibiotic regimens with alternate antibiotic regimens or placebo.^43,64–68 Two trials included data on infants with isolated fast breathing (Afrinest 2015b, Tikmani 2017), 4 included infants with signs of clinical severe infection (Afrinest 2015a, Baqui 2015, Mir 2015, and Zaidi 2012), and no study specifically enrolled infants with critical illness. Detailed individual-level trial characteristics are shown in Table 2. All studies were LMIC-based and limited their study population to infants for whom referral was either refused or not possible. The definition of sepsis in these studies was based on clinical signs of PSBI, though some studies also collected blood cultures.

Two trials included nonhospitalized infants aged 0 to 6 days with fast breathing as the only sign of illness (Tables 2 and 3, Supplemental Tables 5–9 and 16).^64,68 Both trials enrolled infants aged 0 to 59 days but provided disaggregated data for infants aged 0 to 6 days to be included in this review. The results of infants aged 7 to 59 are presented elsewhere.⁶⁹ One trial (Afrinest 2015b) compared PO amoxicillin to IM penicillin plus gentamicin and found a similar likelihood of treatment failure (RR 0.99 [95% CI 0.77 to 1.27], 1 trial, n = 882, very low COE), mortality (RR 1.50 [95% CI 0.25 to 8.93], n = 882, very low COE), and relapse (RR 1.08 [95% CI 0.50 to 2.35], n = 882, very low COE) between groups, whereas the RR of severe adverse events was not estimable because of lack of events in either arm.⁶⁸ This trial did not report on treatment success or neurologic sequelae. The second trial (Tikmani 2017) compared PO amoxicillin to placebo and found a trend to a lower likelihood of treatment failure in the amoxicillin group, but the 95% CI included the null (RR 0.45 [95% CI 0.17 to 1.16], n = 426, low COE), and similar likelihood of relapse (RR 0.68 [95% CI 0.26 to 1.75], n = 210, very low COE) and hospitalization (RR 0.49 [95% CI 0.09 to 2.63], n = 210, very low COE) between groups, whereas the RR was not estimable for mortality because of lack of events.⁶⁴ This trial did not report on outcomes of treatment success, adverse events, or neurologic sequelae.

Four trials of infants with signs of clinical severe infection compared antibiotic regimens that included PO antibiotics versus IM penicillin or gentamicin for 7 days (Tables 2 and 3, Supplemental Fig 5, Supplemental Tables 5–9 and 16).^65–67 Three trials (Afrinest 2015a, Baqui 2015, and Mir 2015) were harmonized to evaluate similar comparator groups. The first comparison, PO amoxicillin plus IM gentamicin × 7 days compared with IM penicillin or gentamicin × 7 days, resulted in a similar likelihood of mortality (RR 0.86 [95% CI 0.55 to 1.34], 3 trials, n = 5054, low COE), treatment failure (RR 0.86 [95% CI 0.72 to 1.02], 3 trials, n = 5054, low COE), relapse (RR 1.05 [0.57 to 1.93], 2 trials, n = 3294, low COE), treatment success (RR 0.99 [95% CI 0.97 to 1.03], 1 trial, n = 1640, low COE), hospitalization (RR 0.78 [95% CI 0.58 to 1.05], 2 trials, n = 3276, low COE), and serious adverse events (RR 1.00 [0.27 to 3.69], 3 trials, n = 5054, low COE) but did not report on neurologic sequelae. For the second comparison, which was only in 1 trial (Afrinest 2015b), PO amoxicillin × 7 days plus IM gentamicin × 2 days compared with IM penicillin or gentamicin × 7 days resulted in lower rates of treatment failure in the intervention group (RR 0.67 [95% CI 0.47 to 0.95], 1 trial, n = 1784, moderate COE) but found similar likelihood of mortality (RR 0.92 [95% CI 0.41 to 2.08], 1 trial, n = 1784, very low COE), relapse (RR 0.85 [95% CI 0.31 to 2.35], 1 trial, n = 1676, very low COE), or serious adverse events (RR 0.33 [95% CI 0.01 to 8.22], 1 trial, n = 1784, very low COE) and did not report on treatment success or neurologic sequelae. For the third comparison, IM penicillin or gentamicin × 2 days followed by PO amoxicillin × 5 days compared with IM penicillin or gentamicin × 7 days demonstrated a similar likelihood of mortality (RR 1.16 [95% CI 0.77 to 1.75], 3 trials, n = 5066, very low COE), treatment failure (RR 0.98 [95% CI 0.81 to 1.19], 3 trials, n = 5066, moderate COE), relapse (RR 0.55 [95% CI 0.16 to 1.97], 2 trials, n = 3329, very low COE), treatment success (RR 1.01 [95% CI 0.98 to 1.05], 1 trial, n = 1639, low COE), hospitalization (RR 0.91 [95% CI 0.69 to 1.22], 2 trials, n = 3276, low COE), and serious adverse events (RR 0.89 [95% CI 0.22 to 3.57], 3 trials, n = 5066, low COE) and did not report on neurologic outcomes.

The fourth trial (Zaidi 2012) evaluated 2 additional simplified antibiotic regimens: (1) PO cotrimoxazole plus IM gentamicin or (2) IM ceftriaxone as compared with IM penicillin plus gentamicin.⁴³ We report the results of the ceftriaxone versus penicillin plus gentamicin comparison above in the pooled analysis of third-generation cephalosporins as compared with penicillin or ampicillin plus gentamicin because it was identical to other regimens encompassed within that analysis. PO cotrimoxazole plus IM gentamicin was associated with a higher likelihood of mortality (RR 5.58 [95% CI 1.26 to 24.72], n = 288, moderate COE) and treatment failure (RR 2.03 [95% CI 1.09 to 3.79], n = 288, moderate COE) but lower likelihood of treatment success (RR 0.90 [95% CI 0.82 to 0.99], n = 288, moderate COE) and similar likelihood of adverse events (RR 5.07 [95% CI 0.25 to 104.67], n = 288, very low COE) compared with IM penicillin plus gentamicin. Zaidi (2012) did not report on the critical outcome of neurologic sequelae.

There were no studies that reported on antibiotic regimens for treatment of critical illness.

This systematic review was conducted to compare the efficacy of antibiotic regimens for the treatment of young infant sepsis. For hospital-based antibiotic regimens, we identified 5 trials comparing third-generation cephalosporins (WHO second-choice antibiotics) to penicillin or ampicillin plus gentamicin (WHO first-choice antibiotics) for which there were similar rates of treatment success. For the nonhospital-based management of sepsis, we identified 6 trials of antibiotic regimens that included oral antibiotics and found similar treatment failure rates between PO amoxicillin plus IM gentamicin and the standard 7-day IM penicillin plus gentamicin course. Overall, there have been limited new RCT data published in the last 10 years.

The landscape of young infant sepsis is evolving in part because of the ever-growing trend of AMR. Recent large global surveillance studies, such as the NeoObs study (11 countries), have demonstrated high rates of resistance to the first-choice WHO drugs, ampicillin and gentamicin, though most of these data were derived from tertiary hospital-based settings and may not be applicable to district hospitals and community-based settings.¹¹ This evidence of AMR is consistent with findings published in other large observational studies, such as Burden of Antibiotic Resistance in Neonates from Developing Societies (BARNARDS) in Bangladesh, Ethiopia, India, Pakistan, Nigeria, Rwanda, and South Africa, which was also conducted in tertiary settings.⁷⁰ 

Clinicians caring for young infants are frequently faced with difficult decisions as they balance the benefits of early treatment based on clinical signs with the risk of propagating resistance patterns. This clinical challenge is compounded in resource-limited settings without access to blood cultures, sensitivity profiles, or local antibiograms. The WHO has made global guidelines to support clinicians by providing recommendations of first-choice and second-choice antibiotic regimens; new guidelines will be published in 2024.^5,14 

Two recent Cochrane reviews evaluated the efficacy of antibiotics for the treatment of EOS and LOS in hospital settings (n = 9 included trials).^16,17 Our search was designed as an update of these reviews, though we also included 26 trials that had been excluded from the previous reviews because of the use of narrow postnatal age or setting limitations. Similar to our findings, most studies were deemed to be at high risk of bias. A third systematic review was conducted as part of the guideline development for the treatment and prevention of neonatal sepsis for the National Institute for Health and Care Excellence guidelines in the United Kingdom. This systematic review was targeting antibiotics used for the treatment of LOS and identified 9 RCTs, all of which were also included in our review.⁷¹ 

No meta-analyses were conducted in the previous systematic reviews because of the heterogeneity of intervention and comparator groups; however, our review had a broader population because we included infants aged 0 to 59 days. As such, we had sufficiently similar comparisons to conduct a small number of meta-analyses.

In our meta-analysis of 5 trials comparing third-generation cephalosporins to penicillin or ampicillin plus gentamicin, we found no difference in treatment failure or success between groups, but we found a better adverse effect profile (including jaundice) among the cephalosporin group. This finding is notable given historic concerns related to cephalosporin use and jaundice in young infants but is consistent with 2 recent studies that did not show an increased risk of hyperbilirubinemia among full term infants treated with ceftriaxone.^72,73 Additionally, a recent systematic review concluded that there is inadequate data to determine the association between ceftriaxone and hyperbilirubinemia given limitations of the current evidence, including small sample sizes, poor external validity, and inconsistency among outcomes ascertained.⁷⁴ However, the potential benefit of third-generation cephalosporins must be weighed with the potential problems with AMR and decreased cost-effectiveness of third-generation cephalosporins as compared with ampicillin or penicillin plus gentamicin. In 2012, the WHO conducted a similar meta-analysis of 4 trials comparing a third generation cephalosporin to ampicillin or penicillin plus gentamicin as part of the systematic review process for the creation of guidelines for the treatment of young infant sepsis.⁷⁵ The 2012 meta-analysis included 3 unpublished trials that were not included in our review because we only included published trials. The previous meta-analysis found similar results to ours with similar mortality and treatment success rates between groups but overall low COE.⁷⁵ 

We found similar rates of treatment failure but lower length of hospital stay in our meta-analysis comparing 10 versus 14 days of antibiotic therapy. These findings are consistent with a recent systematic review comparing antibiotic duration regimens, which included the same studies and reached similar conclusions regarding the low COE and the possibility for a shortened length of stay with similar treatment results.⁷⁶ The shorter length of stay may have potentially important financial implications for families in settings where patients are paying out-of-pocket for healthcare expenditures, as is the case in many LMICs. Furthermore, a shortened hospital stay may be associated with lower exposure to healthcare-associated pathogens and lower risk for the development of AMR.

Three previous systematic reviews or meta-analyses have been conducted to summarize the efficacy of antibiotic regimens for the nonhospital treatment of young infant sepsis. A previous Cochrane systematic review evaluated the efficacy of nonhospital-based antibiotic regimens for treatment of neonatal infections.¹⁵ Similar to our findings, antibiotic regimens that included PO antibiotics were shown to be as effective as standard regimens with regard to treatment failure with the exception of PO cotrimoxazole, which was demonstrated to be less effective. Although the results were similar, the Cochrane review was different from ours in that it included unpublished, disaggregated neonatal (0–28 days) mortality outcome data and did not present certain secondary outcomes, such as treatment success, relapse, and hospitalization, that we included in our review.¹⁵ An individual participant data meta-analysis of the 3 trials (Afrinest 2015a, Baqui 2015, and Mir 2015) examining PO amoxicillin plus IM gentamicin regimens as compared with IM penicillin plus gentamicin found very similar effects on treatment failure to this review; however, their confidence intervals were slightly narrower and did not cross the null, suggesting a benefit of the PO amoxicillin regimens.⁷⁷ A third systematic review of nonhospital studies was conducted in 2015 to inform the creation of the WHO guidelines for the management of possible serious bacterial infection in young infants when referral is not feasible.⁵ This review compared antibiotics for the treatment of fast breathing among infants aged 0 to 59 but did not disaggregate results based on the 0 to 6 day versus 7 to 59 days age groups. They also included the same 4 trials for the treatment of clinical severe infection as our review. They determined COE using the RD per 95% CI as calculated from the PP results reported by the trials, whereas we used the RR as calculated from the ITT results for our determination of COE, per Cochrane handbook. The results from this review were similar to ours except that, when using the PP results, they found treatment failure to be lower with PO amoxicillin plus IM gentamicin for 7 days as compared with IM penicillin plus gentamicin for 7 days (RD −1.8% [95% CI −3.4% to 0.2%], 3 trials, moderate COE).⁵ In our meta-analysis, we found equivalent results using the ITT treatment failure event rates.

The biologic explanation for the potential benefit of PO regimens may be attributable to a few factors. Amoxicillin has a slightly broader spectrum of antimicrobial coverage than penicillin, which may have proven beneficial. Adherence was reported to be higher among those enrolled in the oral regimens, which may have resulted in better outcomes for this group in the ITT analysis.

Overall, the relative efficacy, or even possible benefits, of PO regimens has important implications for settings with limited access to care, as it suggests that young infants with PSBI without signs of critical illness may be appropriately managed with outpatient oral therapies. Importantly, all of these studies included infants with PSBI but excluded infants with signs of critical, severe illness, so trial results for the nonhospital treatment of young infant sepsis is not applicable to this subgroup.

Our systematic review had certain limitations. First, we were inherently limited in our ability to draw conclusions by the quality and heterogeneity of the trials, which were largely small or at high risk of bias. The trials varied widely with regard to their definitions of sepsis and antibiotic regimens. Furthermore, many trials were very old, preceding the modern era of AMR and subsequently lacking relevance in the modern clinical setting. Secondly, as an update to the Cochrane systematic reviews, we were dependent upon the scope of the searches conducted within the previous reviews. Although we reviewed their excluded studies lists, they may have excluded certain studies at the title or abstract screening phase that we could have missed, though we conducted an overview of systematic reviews and bibliography searches to supplement potential missed articles.

We identified 41 trials of antibiotic regimens for the treatment of sepsis in young infants aged 0 to 59 days, 16 of which were published in the last 10 years. There was limited evidence suggesting that WHO first- or second-choice regimens are similar to or superior to alternative regimens, and most estimates were imprecise. Overall, both hospital- and nonhospital-based studies were heterogeneous with respect to interventions, limiting the ability to pool studies. Given the changing global epidemiology of sepsis and AMR profile, our systematic review underscores the need for contemporaneous, adequately powered trials to inform policies related to antibiotic recommendations for young infant sepsis.

We would like to thank Shiyam Sunder Tikmani and the Afrinest team (Simon Cousens) that provided additional analysis for this review question; and Oludare Odumade for her assistance in determining the risk of bias.

AMR

antimicrobial resistance

aOR

adjusted odds ratio

CI

confidence interval

COE

certainty of evidence

EOS

early-onset sepsis

IM

intramuscular

ITT

intention to treat

IV

intravenous

LMICs

low- and middle-income countries

LOS

late-onset sepsis

MD

mean difference

PO

per os/oral

PP

per-protocol

PSBI

possible serious bacterial infection

RCT

randomized controlled trial

RD

risk difference

RR

relative risk

RoB

risk of bias

WHO

World Health Organization