Global incidence and mortality of neonatal sepsis: a systematic review and meta-analysis

Background Neonates are at major risk of sepsis, but data on neonatal sepsis incidence are scarce. We aimed to assess the incidence and mortality of neonatal sepsis worldwide. Methods We performed a systematic review and meta-analysis. 13 databases were searched for the period January 1979–May 2019, updating the search of a previous systematic review and extending it in order to increase data inputs from low-income and middle-income countries (LMICs). We included studies on the population-level neonatal sepsis incidence that used a clinical sepsis definition, such as the 2005 consensus definition, or relevant ICD codes. We performed a random-effects meta-analysis on neonatal sepsis incidence and mortality, stratified according to sepsis onset, birth weight, prematurity, study setting, WHO region and World Bank income level. Results The search yielded 4737 publications, of which 26 were included. They accounted for 2 797 879 live births and 29 608 sepsis cases in 14 countries, most of which were middle-income countries. Random-effects estimator for neonatal sepsis incidence in the overall time frame was 2824 (95% CI 1892 to 4194) cases per 100 000 live births, of which an estimated 17.6% 9 (95% CI 10.3% to 28.6%) died. In the last decade (2009–2018), the incidence was 3930 (95% CI 1937 to 7812) per 100 000 live births based on four studies from LMICs. In the overall time frame, estimated incidence and mortality was higher in early-onset than late-onset neonatal sepsis cases. There was substantial between-study heterogeneity in all analyses. Studies were at moderate to high risk of bias. Conclusion Neonatal sepsis is common and often fatal. Its incidence remains unknown in most countries and existing studies show marked heterogeneity, indicating the need to increase the number of epidemiological studies, harmonise neonatal sepsis definitions and improve the quality of research in this field. This can help to design and implement targeted interventions, which are urgently needed to reduce the high incidence of neonatal sepsis worldwide.


INTRODUCTION
Sepsis is a dysregulated host response to infection leading to life-threatening organ dysfunction. 1 Although neonates are highly vulnerable to sepsis, incidence estimates for this age group are lacking from many countries. 2 Recently, the Global Burden of Disease (GBD) Study 2016/2017 estimated 1.3 (95% CI 0.8 to 2.3) million annual incident cases of neonatal sepsis worldwide, 3 resulting in 203 000 (95% CI 178 700 to 267 100) sepsis-attributable deaths. 4 Neonates are disproportionately affected in low-income and middle-income countries (LMICs) with a high prevalence of infectious diseases 5 and poor access to adequately equipped and staffed healthcare facilities. 6 In sub-Saharan Africa alone, an estimated 5.3-8.7 million disability-adjusted lifeyears have been lost in 2014 due to neonatal sepsis and consecutive long-term morbidity. 7 Neonatal sepsis has resulted in an estimated economic burden of up to US$469 billion in this region (2014 data). 7 In a previous systematic review and meta-analysis, we compiled evidence on the burden of paediatric sepsis including neonatal sepsis. 2 Only eight studies that used a clinical definition of neonatal What this study adds? ► We found in 26 studies a pooled neonatal sepsis incidence of 2824 sepsis cases per 100 000 live births (95% CI 1892 to 4194) and a mortality of 17.6% (95% CI 10.3% to 28.6%). Preterm and very low birthweight neonates were particularly affected, and there were considerable regional differences in incidence. ► Data are lacking from many countries, underlining the need for further epidemiological research to guide interventions that reduce neonatal incidence and mortality.

Original research
Global child health sepsis could be identified, of which five studies originated from LMICs. We aimed to assess the global incidence and mortality of neonatal sepsis with a particular focus on LMICs by updating and extending this previous systematic review. Secondary objectives were to compile data on underlying organisms and antimicrobial resistance, hospital length of stay and sepsis-attributable mortality.

METHODS
The protocol of this systematic review was registered in PROS-PERO (CRD42020149085).

Search strategy and selection criteria
As starting point, we used the previously published systematic review on neonatal sepsis incidence. 2  Studies based on sepsis-relevant ICD-9/ICD-10 codes were also included. Since definitions varied widely, we accepted definitions of early-onset sepsis (EOS) as sepsis within the first 2-7 days of life. Depending on the EOS definition in each study, late-onset sepsis (LOS) was defined as sepsis diagnosed from 3 to 8 days until 28 days after birth, resulting in individual studies with no double counting, but in overlap between the definitions of EOS and LOS. We excluded studies that solely reported culture-proven sepsis cases, incidence per hospital admissions, or used the definition of possible severe bacterial infections (PSBIs) instead of neonatal sepsis, since neonatal sepsis accounts for only one out of four cases of PSBI. 12 We excluded studies limited to other subgroups of sepsis, pathogens or specific patient groups, and studies with a missing study methodology description.

Data compilation and risk of bias assessment
Abstracts and full texts were reviewed by two independent investigators (CF, FR, MT, RH and YS). Discrepancies were resolved by discussion. Non-English articles were assessed by native speakers with medical backgrounds. Data extraction items and strategies can be found in the online supplemental appendix. Pathogen and antibiotic resistance data were only considered from the year 2000 onwards. Authors were contacted to specify methods and to provide additional data. Studies that met the inclusion criteria were assessed for risk of bias by the Hoy tool. 13

Statistical analyses
We categorised the studies by sepsis type (EOS, LOS and EOS/ LOS combined). For subgroup analyses, participants were stratified by birth weight into normal birth weight (≥2500 g), low birth weight (1500-2499 g) and very low birth weight (<1500 g) neonates, and by gestational age into preterm (<37 weeks) and term neonates. Countries were categorised by WHO region and World Bank income level, and studies by communitybased or hospital-based design and decade in which the study was conducted according to the start of their observation period (before 1989, 1989-1998, 1999-2008 and 2009-2018). We conducted meta-analyses of sepsis incidence per 100 000 live births, mortality per 100 sepsis cases and the proportion of culture-proven sepsis with 95% CI using the package 'meta' V.4.9.5 in R V.3.6.1 (online supplemental appendix 2). We calculated pooled estimates using a random-effects model with variance stabilising logit transformed proportions and estimated between-study variance τ² using the Sidik-Jonkman estimator. We used I² statistics to quantify statistical heterogeneity.

RESULTS
We identified 4737 records (figure 1); 250 studies underwent full-text screening; and 26 fulfilled the inclusion criteria.  The entire study population comprised 2 797 879 live births in 14 countries and five WHO regions (online supplemental appendix eTable 1 and figure 2). Two studies originated from low-income countries, 20 from middle-income and four from high-income countries (HICs). Studies were mostly prospective (n=18/26). We included 21 cohort studies, 4 trials and 1 casecontrol study. The majority were single-centre studies. Seven studies used a community-based design; 19 were hospital-based, including studies limited to neonatal intensive care units (NICUs) or studies with entire hospitals under observation. Studies were at moderate to high risk of bias (online supplemental appendix eTable 2).
Overall, 29 608 sepsis cases were identified. Sepsis definitions varied by diagnostic criteria and time intervals (online supplemental appendix Table e1).
Incidence of EOS/LOS combined was reported by 21 studies, of which 5 provided separate results for EOS and LOS. EOS and LOS were distinct subgroups in these studies with no overlap. The remaining five studies reported exclusively on EOS. If not stated otherwise, our results refer to studies that report on EOS/ LOS combined. Separate analyses for EOS or LOS are provided if the number of studies was sufficient.
Birth weight and gestational age were inversely related to EOS incidence, resulting in the highest incidence in VLBW (17 129/100 000, 95% CI 9192 to 29 679) and preterm neonates
We observed substantial heterogeneity between studies, even within subgroup analyses.

Global child health
Klebsiella spp (online supplemental appendix table 3). Results on antimicrobial resistance, which was reported in 3 out of 16 studies, are presented in online supplemental appendix table 3.

DISCUSSION
Our estimates confirm that neonatal sepsis is an important contributor to neonatal morbidity worldwide, with a higher burden in LMICs. Based on the available evidence of 26 studies from 14 countries, we found an incidence of 2824 neonatal sepsis cases per 100 000 live births in the overall time frame. Mortality was 17.6%. The incidence estimate was 1.4-fold higher in the past decade (2009-2018, 3930/100 000 live births), but this estimate was exclusively relying on LMIC data, which might be one reason for the higher incidence. No recent data from HIC could be identified, indicating gaps in epidemiological research on neonatal sepsis all over the world. The overall incidence and mortality for neonatal sepsis was higher than the estimates for EOS and LOS. This could be explained by inclusion of different studies in the three meta-analyses. Incidence of EOS was higher in risk groups, such as LBW, VLBW and preterm infants. In our analyses, the observed population-level incidence of neonatal sepsis was more than four times higher in community-based compared with hospital-based studies. This may reflect high rates of mothers delivering outside a healthcare facility without a skilled birth attendant, unsterile cord care practices, reduced access to healthcare facilities and lower care seeking behaviours 40 in LMICs, where the included communitybased studies originated from. On the other hand, hospitalisation, improved survival of preterm newborns and use of invasive devices increase the incidence of nosocomial LOS cases, especially in HIC. 41 42 Our most recent results of neonatal sepsis incidence in the past decade (2009)(2010)(2011)(2012)(2013)(2014)(2015)(2016)(2017)(2018) indicate an approximately four times higher number of global neonatal sepsis cases than the GBD estimates (3930 vs 937 cases per 100 000), likely due to different methodologies, such as varying case definitions, data collection methods

Original research
Global child health

Original research
Global child health and data inputs from different countries. Extrapolations from our findings must be interpreted with caution since the included studies may not be representative of the global population and carry a moderate to high risk of bias. Contrary to most studies that we included in our review, cases of pneumonia are captured in the GBD modelling of pneumonia as a separate entity, but not as neonatal sepsis cases, 3 which is a major methodological difference. Furthermore, the GBD study used more recent data to compute estimates for the year 2017, while our most recent estimates refer to the past decade (2009)(2010)(2011)(2012)(2013)(2014)(2015)(2016)(2017)(2018). 3 The recent GBD study, which used another methodology to assess sepsisassociated deaths and hospital incidence of sepsis, estimated 20.3 million incident sepsis cases in children aged <5 years per year. 43 The authors did not present specific data on neonatal sepsis. However, given that neonates account for approximately two out of three sepsis cases in children aged <5 years, 17 the estimated incidence is likely higher than those in previous GBD studies.
Approximately one-third of infections were culture-proven with S. aureus and Klebsiella spp as the most common causative pathogens. However, in most studies that reported aetiology, the assessments of causative pathogens were based on only a low number of isolates or originated from single-centre studies. A cautious interpretation of these results is therefore necessary, other systematic reviews with a scope on aetiology and antibiotic susceptibility could compile more evidence. 44 45 Susceptibility testing results were reported too seldomly and may be outdated, and the data are too heterogeneous to draw conclusions.
The extended search strategy identified 17 additional studies from LMICs and 1 additional study from an HIC. In general, systematic reviews that lack information from LMICs may benefit from a broader search strategy but inclusion of specific LMIC country names, and filters for LMIC as proposed by the EPOC group. We included data from seven community-based studies that followed up a large number of newborns in partly rural regions by trained village health workers/teams and are of particular value to better understand the neonatal sepsis epidemiology in remote and rural community settings in low-income countries.
Major knowledge gaps on the population-based epidemiology of neonatal sepsis remain in most countries, particularly, but not exclusively, in LMICs. Reasons for this may be the lack of a robust research infrastructure, formal healthcare systems 46 or prioritisation of other important healthcare issues. Furthermore, several large observational studies that target severe neonatal infections did not meet our inclusion criteria either due of their scope or the inclusion criteria.
There are limitations of our study. First, the studies we included may not be representative of the global population for several reasons. Data inputs were from 14 countries from five WHO regions; thus information on the majority of countries and the European WHO region was not included. Numerous studies only reported the incidence of NICU-treated or hospital-treated neonatal sepsis, which may underestimate the incidence of neonatal sepsis by excluding diseased infants outside the hospital setting. This is of particular relevance in LMICs, where a considerable number of births occur outside the hospital 47 and healthcare seeking for sick neonates may be low. 40 Furthermore, most hospital-based studies were performed in tertiary care hospitals, which deliver higher quality of care compared with community hospitals, where a large number of neonates may seek care in some contexts. Second, we cannot judge on temporal trends as the number of studies in the respective time strata was low. Thus, differences in incidence estimates depending on the years of observation may be also explained by the fact that studies originated from different regions and used study designs with limited comparability. Third, the included studies were at moderate to high risk of bias, mainly due to the aforementioned lack of representativeness and a reliable case definition. Fourth, we observed high between-study heterogeneity, possibly due to differing study designs, settings and observation periods. The pooled estimates include population-level estimates from community-based studies, as well as NICU-based and hospital-based studies from facilities of different levels of care. General access, capacities and admission policies may differ between NICUs and countries. Heterogeneity may also be driven by different definitions. Applied sepsis definitions are mostly adaptations of the clinical consensus criteria, 9 such as when laboratory testing was unavailable, 31 or relied on selected clinical symptoms of neonatal sepsis. Currently, there is no sepsis definition for the neonatal age group in line with the current sepsis definition, 48 which can be applied in all resource settings and also in preterm neonates. That leads to wide variations of existing neonatal sepsis definitions 49 and hampers the comparability of sepsis epidemiology studies. In our review, we accepted a wide range of definitions, including those based on clinical criteria, to not exclude studies from LMICs. We pooled estimates from all included studies to gather all available evidence on the epidemiology of neonatal sepsis. Fifth, the underlying causes of death in these neonates were not reported except in one of the included studies, which concluded a sepsisattributable mortality rate of 24%. The remaining studies referred to the proportion of patients with neonatal sepsis who died, but the cause of death was not specified. Finally, while our systematic review and pooled estimates consider only live births, globally, one out of four stillbirths are attributed to infection, 50 adding substantially to the overall burden of perinatal deaths that are due to sepsis.
The limitations identified by our study highlight the need to increase and improve epidemiological research on neonatal sepsis. A crucial prerequisite is the establishment of a harmonised definition and validated diagnostic criteria for neonatal sepsis applicable in all resource settings and in preterm neonates. Furthermore, a standardisation of study designs and reporting, for example, in line with the Strengthening the Reporting of Observational Studies in Epidemiology for Newborn Infection checklist, is important. 51 Further investments in epidemiological research infrastructure and capacity in both LMICs and HICs are needed in order to improve surveillance of neonatal sepsis, particularly in communitybased designs. Having robust methodologies to measure the burden of neonatal sepsis will be extremely useful to assess the effect of urgently needed interventions to prevent it. Indeed, a lot needs to be done yet to prevent infections in neonates and to reduce the frequency of their evolution to sepsis complications. The WHO issued recommendations on key prevention and control measures both in the community and hospital setting and improved early recognition and timely and appropriate treatment of neonatal sepsis, as well as recommendations for early essential newborn care. [52][53][54][55] This includes, among others, exclusive breast feeding, skin-to-skin contact with the mother from birth, the empirical antibiotic treatment of neonates with signs of severe infection/sepsis (eg, fast breathing) and education of families on the recognition of signs of neonatal sepsis by trained community health workers. 54 55 Given that approximately one quarter of neonatal deaths is related to infection and sepsis, progress in reducing the burden of neonatal sepsis is critical to reduce global child mortality and of high public health relevance.
Contributors CF, FR, AC, BA, NK, KR and TE designed the study. CF, FR, RH, MT and YS conducted the literature review, and performed data extraction and the risk of bias assessment. FR performed the statistical analysis. PS and RM advised on the statistical analyses. CF and FR drafted the manuscript. All authors revised the review for important intellectual content and contributed to the writing of the manuscript.

Funding
The WHO provided funding for this study and contributed to the study design, data extraction and interpretation, as well as writing of the manuscript.

Disclaimer
The opinions expressed in this Article are those of the authors and do not reflect the official position of WHO. WHO takes no responsibility for the information provided or the views expressed in this article.

Map disclaimer
The depiction of boundaries on this map does not imply the expression of any opinion whatsoever on the part of BMJ (or any member of its group) concerning the legal status of any country, territory, jurisdiction or area or of its authorities. This map is provided without any warranty of any kind, either express or implied.
Competing interests CF received grants from the German Federal Ministry of Education and Research (BMBF) via the Center for Sepsis Control and Care (FKZ: 01EO1002 and 01EO1502) and from the Innovation Funds of the German Government (FKZ 01VSF17010). The other authors declared no competing interests.

Patient consent for publication Not required.
Provenance and peer review Not commissioned; externally peer reviewed.
Data availability statement Data are available upon reasonable request. The datasets analysed during the current study are available from the corresponding author on reasonable request. The software codes used during the current study are available in online supplemental file 2.
Supplemental material This content has been supplied by the author(s). It has not been vetted by BMJ Publishing Group Limited (BMJ) and may not have been peer-reviewed. Any opinions or recommendations discussed are solely those of the author(s) and are not endorsed by BMJ. BMJ disclaims all liability and responsibility arising from any reliance placed on the content. Where the content includes any translated material, BMJ does not warrant the accuracy and reliability of the translations (including but not limited to local regulations, clinical guidelines, terminology, drug names and drug dosages), and is not responsible for any error and/or omissions arising from translation and adaptation or otherwise.

Open access
This is an open access article distributed in accordance with the Creative Commons Attribution 4.0 Unported (CC BY 4.0) license, which permits others to copy, redistribute, remix, transform and build upon this work for any purpose, provided the original work is properly cited, a link to the licence is given, and indication of whether changes were made. See: https:// creativecommons. org/ licenses/ by/ 4. 0/.