Global Management of Serious Bacterial Infections in Young Infants Aged 0 to 59 Days: An Overview of Systematic Reviews

We used standard Cochrane methods for “Overviews of systematic reviews”¹² to search for all published systematic reviews. We searched for systematic reviews of both RCTs and observational studies. The review was registered in PROSPERO (CRD42023449856).¹⁵ 

The search strategy included terms for infants, SBI, and systematic reviews only; to avoid losing data, we did not restrict the search terms on intervention type or outcome. The search was for the last 5 years; ie, from January 1, 2018 to November 3, 2023. We searched the following databases: MEDLINE via Ovid; Embase via Ovid; the Cochrane Database of Systematic Reviews; the Cochrane Database of Systematic Review Protocols; Epistemonikos; and the PROSPERO International prospective register of systematic reviews. We used the Covidence systematic review software, Veritas Health Innovation, Melbourne, Australia,¹⁶ to manage all stages.

The population included was infants aged 0 to 59 days with clinical syndromes of confirmed and suspected SBI, sepsis, pneumonia, or meningitis, as defined by the study authors, in any setting. Other infections, such as urinary tract infections, were not included.

Intervention and comparators included were (1) diagnostic accuracy (sensitivity and specificity) of 2 or more clinical signs (history or examination) with or without laboratory markers compared with a reference standard (clinician judgement, culture proven sepsis, or all cause or cause specific mortality); and (2) any currently recommended antibiotic regimens compared with different antibiotic regimens.

Outcomes included were all cause or cause specific mortality; morbidity (treatment failure, treatment success); health service use (eg, length of hospital stay, intensive care treatment); neurodevelopmental impairment or disability; and adverse events (eg, ototoxicity) at latest follow up. We excluded any reviews that did not include clinical outcomes; eg, if they reported only on pharmacokinetic outcomes or nonclinical outcomes, such as blood test results. We assessed if studies were conducted in hospital or non-hospital community settings but did not prespecify any subgroup analyses by setting.

Four review authors (G.W., D.S., N.S., K.E.) independently screened titles and abstracts and assessed the full texts of all identified systematic reviews for eligibility. We assessed the reviews’ objectives and methods, including outcomes and participants for relevance, and included only those reviews that met the criteria listed above. We resolved any disagreements through discussion until we reached a consensus.

We generated a data extraction form and pretested it. After verification, 2 review authors (G.W., D.S.) independently extracted data from each review. We resolved any discrepancies through discussion until we reached a consensus or, if necessary, by consulting another review author.

We collected basic data on:

1. review details: title, authors, year of publication, date of last assessment, number of included studies and participants, conflict of interest of review authors, funding details of review, risk of bias and certainty of evidence tools, funding sources and conflicts of interest of included authors of the review;

2. characteristics of review: facility type, geographic location, country, characteristics of the participants: gestational age, sex, birth weight, ethnicity, intervention and comparators;

3. type of antibiotic regimen;

4. diagnostic accuracy (sensitivity and specificity) of clinical signs (history, examination, laboratory) compared with a reference standard (clinician judgement, culture proven sepsis, or all cause or cause specific mortality);

5. outcomes: mortality, morbidity, neurodevelopment, adverse events at latest follow up.

Two review authors (G.W., D.S.) assessed the methodological quality of the systematic reviews using AMSTAR 2 (A MeaSurement Tool to Assess systematic Reviews 2)¹⁷ and ROBIS (Risk of Bias in Systematic Reviews)¹⁸ tools. We resolved any discrepancies through discussion until we reached a consensus.

AMSTAR 2 assesses the degree to which review methods avoid bias by evaluating the methods against specific criteria.¹⁷ There are 7 critical domains: protocol registered before commencement of the review (item 2); adequacy of the literature search (item 4); justification for excluding individual studies (item 7); risk of bias from individual studies being included in the review (item 9); appropriateness of meta-analytical methods (item 11); consideration of risk of bias when interpreting the results of the review (item 13); and assessment of presence and likely impact of publication bias (item 15).¹⁷ AMSTAR 2 has 4 final ratings of “overall confidence in the review”: (1) critically low (more than 1 critical flaw with or without noncritical weaknesses; ie, the review has critical flaws and may not provide an accurate and comprehensive summary of the available studies that address the question of interest); (2) low (1 critical flaw with or without noncritical weaknesses; ie, the review has a critical flaw and may not provide an accurate and comprehensive summary of the available studies that address the question of interest); (3) moderate (more than 1 noncritical weakness; ie, the systematic review has more than 1 weakness but no critical flaws; it may provide an accurate summary of the results of the available studies that were included in the review); and (4) high (no or 1 noncritical weakness; ie, the systematic review provides an accurate and comprehensive summary of the results of the available studies that address the question of interest).

The ROBIS tool has 3 categories or “phases”.¹⁸ Phase 1 assesses the relevance of the review. Phase 2 assesses the review process and has 4 domains: study eligibility, identification and selection of studies, data collection and study appraisal, and synthesis of findings. The final phase assesses the overall risk of bias based on the interpretation of review findings and limitations. ROBIS has 3 final risk of bias ratings: (1) low risk of bias (the findings of the review are likely to be reliable, there are no concerns with the review process or concerns were appropriately considered in the review conclusions, and the conclusions were supported by the evidence and included consideration of the relevance of included studies); (2) high risk of bias (1 or more of the concerns raised during the assessment was not addressed in the review conclusions, the review conclusions were not supported by the evidence, or the conclusions did not consider the relevance of the included studies to the review question); and (3) unclear risk of bias (there is insufficient information reported to make a judgement on risk of bias).

We described the data narratively. We described each systematic review by disease syndrome, interventions, and age group. We also assessed the methodological quality of the reviews using AMSTAR 2 and ROBIS criteria and the overall total quality rating.

We also assessed whether each systematic review assessed antibiotic regimens listed in the WHO EML for Children “fully” (all 4 components of the PICO including our age group of interest 0–59 days), “partially” (1 to 3 components of the PICO), or “did not address” (ie, addressed no components of the PICOs).

We identified 3004 records and screened 2409 titles and abstracts after duplicates were removed. 212 studies were assessed for eligibility with 194 studies excluded with reasons provided in Fig 1. Nine reviews were included in this Overview (Persad 2021,¹⁹ Sofouli 2022,²⁰ Duby 2019,²¹ Korang 2021b,²² Korang 2021c,²³ NICE 2021,²⁴ Korang 2021a,²⁵ Aljarbou 2023,²⁶ Van Hentenryck 2022²⁷).^19–27 

Of the 194 excluded studies, 28 reviews were excluded because of wrong population (eg, not sepsis, pneumonia, or meningitis), 46 had the wrong study design (eg, were literature reviews without a systematic review or meta-analysis), 116 had the wrong intervention (eg, immunomodulator not antibiotic), 4 had the wrong outcomes (eg, pharmacokinetic outcomes only). Characteristics of excluded reviews can be found in Appendix 2 (Supplemental Information).

The characteristics of the 9 included reviews are shown in Table 2. There were a total of 125 studies and 22 806 infants in the 9 reviews.

Seven reviews included RCTs only (Sofouli 2022, Duby 2019, Korang 2021b, Korang 2021c, NICE 2021, Korang 2021a, Aljarbou 2023) and 2 included observational studies as well as RCTs (Persad 2021, Van Hentenryck 2022). There were no reviews of network meta-analyses. The number of studies in the reviews ranged from 4 (Korang 2021a) to 36 (Persad 2021) (median 10, interquartile range 5 to 16, mean 14, SD 12). Three reviews (Persad 2021, NICE 2021, Duby 2019) did not report the total number of participants. The number of participants in the other 6 reviews ranged from 84 (Korang 2021a) to 18 096 (Persad 2021) (median 580, interquartile range 266 to 1765, mean 3258, SD 6602).

All reviews included studies from LMICs, 7 reviews included studies from high income countries. Duby 2019 specifically excluded high income countries and Aljarbou 2023 located LMICs studies only.

One review searched for hospitalized infants only (Korang 2021a), 1 searched for community-based care only (Duby 2019), and the remainder searched for both hospitalized and nonhospitalized infants (Persad 2021, Korang 2021b, Korang 2021c, NICE 2021). After final screening, 7 reviews included hospital studies only, and 1 review included community studies only (Duby 2019). One review did not describe the setting (Sofouli 2022). No reviews included both hospital and community-based studies.

Two reviews examined diagnostic accuracy for sepsis (Persad 2021, Sofouli 2022), no reviews examined diagnostic accuracy for pneumonia or meningitis. Four reviews examined antibiotic effectiveness for sepsis (Duby 2019, Korang 2021b, Korang 2021c, NICE 2021), 1 review examined antibiotic effectiveness for pneumonia (Korang 2021a), and no reviews examined antibiotic effectiveness for meningitis. One review examined antibiotic duration for sepsis (Aljarbou 2023), 1 review examined antibiotic duration for meningitis (Van Hentenryck 2022), and no reviews examined antibiotic duration for pneumonia.

All reviews met the PICOs “partially” (1 to 3 components of the PICO), but none met our PICOs fully (all 4 components of the PICO for infants aged 0–59 days).

The 2 reviews that examined diagnostic accuracy for sepsis (Persad 2021, Sofouli 2022) included only specialized clinical decision support algorithms and predictive modeling tools and excluded all other algorithms, such as simple Integrated Management of Childhood Illness type check lists.

Three of the 4 reviews that examined antibiotic effectiveness for sepsis (Duby 2019, Korang 2021b, Korang 2021c, NICE 2021) assessed selected specialized populations, 1 assessed early onset sepsis only (Korang 2021b), 1 assessed late onset sepsis only (Korang 2021c), and 1 assessed community-based studies only (Duby 2019). The other review (NICE 2021) assessed English language studies only. No studies reviewed both hospital and community-based studies; ie, studies across all settings and across our target age group of 0 to 59 days.

The only review that examined antibiotic effectiveness for pneumonia (Korang 2021a) was designed to assess hospital-based studies only. No reviews were located that assessed both hospital and community-based studies.

The only review that examined antibiotic duration for sepsis (Aljarbou 2023) searched for all studies conducted in infants aged 0 to 28 days, not our target age group of 0 to 59 days. This review was also not designed to assess other antibiotic regimens, such as comparisons of different types of antibiotics “head-to-head”.

The only review that examined antibiotic duration for meningitis in infants aged 0 to 90 days (Van Hentenryck 2022) was not designed to assess other antibiotic regimens, such as comparisons of different types of antibiotics “head-to-head”.

Eight reviews used the Cochrane Risk of Bias Tool Version 2 (Cochrane RoB 2), 1 used the Cochrane Risk of Bias in Non Randomized Studies of Interventions (Cochrane-ROBINS-I) (Persad 2021), and 1 used the Newcastle-Ottawa Quality Assessment Scale for Observational Studies and the Johanna Briggs Institute scale for case series (Van Hentenryck 2022). One review (Sofouli 2022) did not assess risk of bias.

Six reviews used Grading of Recommendations Assessment, Development and Evaluation (GRADE)²⁸ to assess the certainty of the body of evidence, and 2 (Sofouli 2022, Van Hentenryck 2022) did not assess certainty of evidence. NICE 2021 used a modified form of GRADE where imprecision was not used to downgrade outcome certainty.

The “AMSTAR 2” quality assessment is summarized in Table 3 and full details are in Appendix 4 (Supplemental Information). Two reviews (Van Hentenryck 2022, Sofouli 2022) were graded as “critically low confidence in results” by AMSTAR 2. Van Hentenryck 2022 was rated down for 2 critical items: item 4 (comprehensive search) as the authors did not appear to justify their publication restrictions (ie, the authors restricted the search only to English but did not explain why), and item 7 (excluded studies justification) as the authors did not provide a list of excluded studies or justify the exclusions. Sofouli 2022 was rated down for the same 2 critical items as Van Hentenryck 2022 and in addition was rated down for item 1 (a priori protocol registration), ie, there was no a priori registration of the protocol. In addition, for item 4 (comprehensive search), Sofouli 2022 only searched 1 database (PubMed). In addition, for item 9 (ROB assessment) Sofouli 2022 did not state whether they assessed risk of bias, and for item 13 (ROB in individual studies) the authors did not state if they considered the risk of bias in the individual studies when drawing their final conclusions.

Three reviews (NICE 2021, Persad 2021, Aljarbou 2023) were graded as “low confidence in results” by AMSTAR 2. NICE 2021 was rated down for item 4 (comprehensive search) as the authors did not appear to justify their publication restrictions, ie, the authors restricted the search only to English but did not explain why. Also, for a noncritical domain (item 6 - data extraction), it was not possible to confirm from the paper or the methods documents whether there had been double data extraction or individual extraction with checked agreement. Persad 2021 was rated down for item 7 (excluded studies justification) as the authors also did not provide a list of excluded studies or justify their exclusions. Aljarbou 2023 was rated down for item 11 (statistical methods) as the authors did not justify their meta-analysis method. Four studies (Duby 2019, Korang 2021a, Korang 2021b, Korang 2021c) were graded as “high confidence in results” by AMSTAR 2, these were all Cochrane reviews.

The “ROBIS” quality assessment is summarized in Table 3 and full details are in Appendix 5 (Supplemental Information). Three reviews (Sofouli 2022, NICE 2021, Van Hentenryck 2022) were graded as in AMSTAR 2 with concerns. Van Hentenryck 2022 was graded as high risk of bias for 2 domains; for domain 1 (study eligibility criteria) as the included studies were restricted based on language (English) with no justification provided; for domain 2 (identification and selection of studies) as the search strategies were restricted to English and it was unclear whether additional search methods (eg, reference checking) had been performed. NICE 2021 was rated down for domain 1 and domain 2 for the same reasons as Van Hentenryck 2022. In NICE 2021 it was also unclear whether the remaining titles and abstracts were screened by a single reviewer or by machine learning. Sofouli 2022 was rated down for the same reasons as NICE 2021 for domains 1 and 2. In addition, Sofouli 2022 was rated down further for domains 1 to 3 as there was no a priori published protocol or registration, and it was not possible to assess the study methods and analysis against a priori decided methods. Also, Sofouli 2022 appeared to search only 1 database (PubMed). For domain 3 (data collection and study appraisal), no information was provided on the data collection process (eg, double data extraction). No details of assessment of risk of bias were provided, so it was not possible to appraise the study against predefined criteria for domain 4 (synthesis and findings). Also, as there was no protocol, it was not possible to assess the synthesis against the desired plan. However, the conclusion of Sofouli 2022 did frame the review as exploratory, stating that more research and standardization is required.

The remaining 6 reviews (Duby 2019, Korang 2021a, Korang 2021b, Korang 2021c, Persad 2021, Aljarbou 2023) were graded by ROBIS as having low risk of bias. Four of these (Duby 2019, Korang 2021a, Korang 2021b, Korang 2021c) were also graded by AMSTAR 2 as “high confidence in results,” but 2 were graded by AMSTAR 2 as “low” (Aljarbou 2023) and “critically low” confidence (Persad 2021).

Despite emerging AMR and the high burden of SBI in infants aged 0 to 59 days in LMICs, we found no reviews that met our 4 PICOs for diagnostic accuracy and antibiotic management of sepsis, pneumonia, and meningitis in the previous 5 years. We also found no other overviews, umbrella, or scoping reviews published over that time that met our PICOs.

We located 9 systematic reviews that had been published in the last 5 years: 2 examined diagnostic accuracy for sepsis (Persad 2021,¹⁹ Sofouli 2022²⁰); 4 reviews examined antibiotic effectiveness for sepsis (Duby 2019,²¹ Korang 2021b,²² Korang 2021c,²³ NICE 2021²⁴); 1 review examined antibiotic effectiveness for pneumonia (Korang 2021a²⁵); 1 review examined antibiotic duration for sepsis (Aljarbou 2023²⁶), and 1 review examined antibiotic duration for meningitis (Van Hentenryck 2022²⁷). However, all reviews examined specialized populations restricted by age (early onset, late onset, neonates) and setting (hospital and community only), or they addressed only highly specialized interventions, such as predictive modeling. None of the 9 reviews examined both hospital and community-based studies; ie, studies across all settings and across our target age group of 0 to 59 days.

The 9 reviews were also of variable quality. There are 3 tools recommended by Cochrane Collaboration to assess the quality of systematic reviews,^17,18,29 of which 2 are widely used (AMSTAR 2¹⁴ and ROBIS¹⁵). AMSTAR 2 and ROBIS consider complementary domains: ROBIS examines bias in detail, whereas AMSTAR 2 assesses bias but also other aspects of quality, such as a funding source and assessments of conflicts of interest. We used both tools in our Overview and found there was marked consistency between the AMSTAR 2 and ROBIS quality ratings.

Four reviews (Duby 2019, Korang 2021a, Korang 2021b, Korang 2021c) were graded by both AMSTAR 2 and ROBIS as having low risk of bias. Three reviews (Sofouli 2022, NICE 2021, Van Hentenryck 2022) were graded by both AMSTAR 2 and ROBIS as having high risk of bias. AMSTAR 2 rated 2 other reviews as high risk of bias and low confidence in results (Persad 2021, Aljarbou 2023) because of critical items (not justifying excluded studies [Persad 2021], and not explaining statistical methods [Aljarbou 2023]), these are domains not included in ROBIS. Some authors have suggested AMSTAR 2 is nondiscriminatory and overdiagnoses bias and poor quality.³⁰ However, we consider that the 2 additional AMSTAR 2 domains (justification of exclusions and description of statistical methods) are important forms of quality assessment.

Our study had some limitations. Our inclusion criteria were limited in scope. Our a priori objective was to understand the gaps in systematic reviews and not to assess effectiveness or report effect estimates. We also only considered reviews published in the last 5 years as we wanted to review contemporaneous data to inform WHO guideline development. Our objectives also were to assess antibiotic regimens for the diagnosis and therapy of SBIs. We did not include preventive interventions, such as vaccines, or other therapies, such as inotropic support, or immune stimulators, such as corticosteroids. The strengths of our review were our rigorous search strategy across multiple databases, our detailed extraction by 2 independent reviewers, and our quality assessment using both AMSTAR 2 and ROBIS. We consider our methods were rigorous and minimized bias.

There are many current publications of the “estimates” of the prevalence of SBI and the “risks” and burden of infants with SBI. However, to our knowledge, this is the first overview of systematic reviews of antibiotic regimens to improve outcomes in infants with SBI. We found no reviews on diagnostic accuracy and therapies for sepsis, pneumonia, or meningitis in the last 5 years that met our PICOs. In particular, few reviews and studies targeted antibiotics in the WHO AWaRe antibiotic classification system or WHO EML for Children. The gaps in rigorous systematic review evidence are concerning as there have been over 150 trials evaluating SBI in LMICs, the field is rapidly evolving in diagnostics and therapies, and there are many concerns about AMR reducing antibiotic efficacy across LMICs.

In 2014, a global research priority setting exercise identified many research gaps for the care of SBI.⁸ Important questions about the efficacy and effectiveness of interventions for SBI have been answered over the last 10 years,³¹ yet our overview has shown that important gaps in the global evidence base also remain for antibiotic efficacy.⁸ Despite emerging AMR, few studies have examined antibiotic regimens currently being used in LMICs and quality is of concern in many. More high quality data are needed to inform management of newborns with the highest burden of ill health, such as those with SBI in LMICs.: