Neonatal sepsis and mortality in low-income and middle-income countries from a facility-based birth cohort: an international multisite prospective observational study

1Neonatal sepsis is a primary cause of neonatal mortality within low-income and middle-income countries (LMICs), with LMICs bearing the burden of 99% of global neonatal mortality, highlighting global disparity., Without significant reduction of infection-related neonatal deaths in LMICs it is unlikely that Sustainable Development Goal 3, which aims to reduce neonatal mortality to at least 12 per 1000 livebirths by 2030 will be met.

2Population-level estimates of laboratory-confirmed sepsis in high-income countries are well studied and documented; however, accurate incidence and associations remain decidedly understudied and undetermined in LMICs., Existing research has identified low birthweight, preterm birth (<37 weeks’ gestation), premature rupture of membranes (PROM), neonatal sex, intrapartum-related complications such as perinatal asphyxia, low socioeconomic status, poor sanitation, malnutrition, and overcrowding as associated with an increased risk of neonatal sepsis.,,, In resource-poor settings, efficient and effective diagnosis of sepsis, including identification and antibiotic susceptibility reporting, is challenging. To date, no studies supported by microbiological data measure incidence of neonatal sepsis and subsequent mortality within a facility-based cohort. The absence of these data jepordises effective intervention development., To further exacerbate the issue, the global rise in multidrug-resistant bacteria, particularly Gram-negative bacteria, compromises antimicrobial stewardship and increases neonatal mortality and morbidity.,,

3Evidence before this study We searched PubMed and Science Direct for articles published in English between Jan 1, 2008, and Dec 31, 2021, using the keywords “neonatal sepsis”, “low- and middle-income countries”, “developing countries”, “Africa”, and “Asia”. Our search strategy yielded 1428 articles of which 72 were relevant. After further scrutiny and excluding reviews, meta-analyses, clinical trials, and focusing on specific cohorts (eg, preterm neonates or those with fungal sepsis) a total of 14 were included. WHO and Save the Children US webpages were consulted to seek additional information. Included articles were examined for richness of epidemiological data, type of institution that conducted the research (public or private institution), single or multisite, definition of sepsis (clinically suspected or laboratory confirmed), and study design. Previous studies on neonatal sepsis focused exclusively on neonatal units within hospitals or are single-site studies. No studies enrolled from a facility-based birth cohort with a denominator of all births. Most studies (12 [92%] of 13) enrolled participants with clinical signs of sepsis from a neonatal intensive care unit and 85% (11 of 13) included laboratory confirmation of sepsis. NeoAMR, ANISA, and the DeNIS study are large neonatal sepsis studies that examined antimicrobial resistance. Both DeNIS and NeoAMR recruited from neonatal intensive care units, and ANISA recruited from the community. NeoOBS study has a similar study design to BARNARDS; however, it is looking at sepsis cases only. Added value of this study BARNARDS is the first prospective observational study looking at the incidence and associations of neonatal sepsis from a facility-based birth cohort in low-income and middle-income countries (LMICs) supported by positive blood cultures with an aim to reduce the gap in research and data in these areas of high burden. BARNARDS assessed data from four domains, health care, maternal, birth and neonatal, and living environment and explored associations with neonatal sepsis and all-cause mortality. BARNARDS enrolled mothers and neonates from 12 facilities in seven LMICs and excluded wholly private health-care institutions, with the intention of capturing areas of low socioeconomic status often precluded in previous studies. Some of the clinical sites had not participated in many research activities, all of which were representative of usual clinical settings reflective of their region. This study provided a characterisation of the burden of neonatal sepsis and mortality in under-researched populations across several LMICs. The longitudinal nature of BARNARDS supported time-to-event analysis with measures of time to clinically suspected sepsis; time to laboratory-confirmed sepsis; and time to all-cause mortality in neonates with no clinically suspected sepsis and no laboratory-confirmed sepsis, clinically suspected sepsis but no laboratory-confirmed sepsis, and clinically suspected sepsis with laboratory-confirmed sepsis (a sub-set of clinically suspected sepsis). Implications of all the available evidence Neonatal sepsis and its associated mortality are key global health concerns that are challenging to manage, especially in LMICs. The international network of BARNARDS provides a synergistic blend of clinical, epidemiological, and microbiological data, generating valuable evidence of neonatal sepsis and mortality following sepsis in LMICs.

4Evidence before this study

5We searched PubMed and Science Direct for articles published in English between Jan 1, 2008, and Dec 31, 2021, using the keywords “neonatal sepsis”, “low- and middle-income countries”, “developing countries”, “Africa”, and “Asia”. Our search strategy yielded 1428 articles of which 72 were relevant. After further scrutiny and excluding reviews, meta-analyses, clinical trials, and focusing on specific cohorts (eg, preterm neonates or those with fungal sepsis) a total of 14 were included. WHO and Save the Children US webpages were consulted to seek additional information. Included articles were examined for richness of epidemiological data, type of institution that conducted the research (public or private institution), single or multisite, definition of sepsis (clinically suspected or laboratory confirmed), and study design. Previous studies on neonatal sepsis focused exclusively on neonatal units within hospitals or are single-site studies. No studies enrolled from a facility-based birth cohort with a denominator of all births. Most studies (12 [92%] of 13) enrolled participants with clinical signs of sepsis from a neonatal intensive care unit and 85% (11 of 13) included laboratory confirmation of sepsis. NeoAMR, ANISA, and the DeNIS study are large neonatal sepsis studies that examined antimicrobial resistance. Both DeNIS and NeoAMR recruited from neonatal intensive care units, and ANISA recruited from the community. NeoOBS study has a similar study design to BARNARDS; however, it is looking at sepsis cases only.

6Added value of this study

7BARNARDS is the first prospective observational study looking at the incidence and associations of neonatal sepsis from a facility-based birth cohort in low-income and middle-income countries (LMICs) supported by positive blood cultures with an aim to reduce the gap in research and data in these areas of high burden. BARNARDS assessed data from four domains, health care, maternal, birth and neonatal, and living environment and explored associations with neonatal sepsis and all-cause mortality. BARNARDS enrolled mothers and neonates from 12 facilities in seven LMICs and excluded wholly private health-care institutions, with the intention of capturing areas of low socioeconomic status often precluded in previous studies. Some of the clinical sites had not participated in many research activities, all of which were representative of usual clinical settings reflective of their region. This study provided a characterisation of the burden of neonatal sepsis and mortality in under-researched populations across several LMICs. The longitudinal nature of BARNARDS supported time-to-event analysis with measures of time to clinically suspected sepsis; time to laboratory-confirmed sepsis; and time to all-cause mortality in neonates with no clinically suspected sepsis and no laboratory-confirmed sepsis, clinically suspected sepsis but no laboratory-confirmed sepsis, and clinically suspected sepsis with laboratory-confirmed sepsis (a sub-set of clinically suspected sepsis).

8Implications of all the available evidence

9Neonatal sepsis and its associated mortality are key global health concerns that are challenging to manage, especially in LMICs. The international network of BARNARDS provides a synergistic blend of clinical, epidemiological, and microbiological data, generating valuable evidence of neonatal sepsis and mortality following sepsis in LMICs.

10The Burden of Antibiotic Resistance in Neonates from Developing Societies (BARNARDS) study was established in 2015 to determine the incidence of and factors associated with neonatal clinically suspected sepsis, laboratory-confirmed sepsis, and all-cause mortality across 12 sites in seven LMICs (Bangladesh, Ethiopia, India, Pakistan, Nigeria, Rwanda, and South Africa; appendix p 2). There were two recruitment pathways: a facility-based birth cohort and neonates admitted to hospital with signs of sepsis. This Article focuses on the facility-based birth cohort and aims to provide an understanding of the extent to which neonatal sepsis and subsequent mortality occur across several LMICs in Africa and south Asia. It also aims to identify the factors associated with both neonatal sepsis and all-cause mortality.

11BARNARDS was a large multisite international prospective observational study, recruiting mothers and neonates aged 0–60 days from 12 clinical sites at large public hospitals located in Rwanda, Bangladesh, Ethiopia, Nigeria, Pakistan, India, and South Africa (appendix p 2) between Nov 12, 2015 and Feb 1, 2018. Some sites included community hospitals, providing urban and rural comparisons (appendix p 3). Ethical approval was obtained at each clinical site (appendix p 4). Mothers were provided with study information in local languages. Informed consent (written, if possible, if not, oral) was collected from the mothers by trained researchers. Mothers presenting to BARNARDS’ clinical sites in labour and their respective liveborn neonates were eligible to be enrolled. Stillborn neonates and their respective mothers were ineligible and excluded if inadvertently included.

12Demographic and clinical data were collected by standardised electronic or paper-based questionnaires if the internet connection posed issues, or staff safety was compromised (appendix pp 5–8). The questionnaires facilitated the analysis of factors associated with neonatal sepsis and all-cause mortality within four domains: health care, maternal, birth and neonatal, and living environment (panel). Telephone or home-visit follow-ups were conducted at days 3, 7, 14, 28, and 60 of life of neonates by research nurses. Neonates were followed up until 60-days old, loss to follow-up, or death (appendix p 9). All sites were provided with a standardised list of symptoms indicative of neonatal sepsis (appendix p 9). Microbiology methods are published elsewhere. Health-care domain (researcher reported) Type of ward MaternityNeonatalObstetrics and gynaecologyOther Number of beds on the ward 1–34–1011–2021 or more Bathroom on the wardMother's location on the ward First third (closest to door)Middle thirdLast third (furthest from door)Other Maternal domain (mother reported) Mother's age 20 years or younger21–35 years36 years or older Pregnancy history First pregnancyPrevious pregnancy did not result in a livebirthLive previous birth Health conditions DiabetesHypertensionImmune-compromisedMalariaTuberculosisOther infectionTyphoidOther health condition Received private health care in previous 3 monthsVisited traditional healer in previous 3 monthsHospitalised in previous 12 monthsUsed antibiotics in previous 3 monthsMother's educational status None or primarySecondary schooling or university Birth and neonatal domain Gestational age, mother-reported Term (37–41 weeks)Preterm (<37 weeks)Post-term (>41 weeks) Premature rupture of membranes, mother reported or researcher reportedDelivery type, researcher reported Natural (spontaneous vaginal delivery)Planned caesarean sectionEmergency caesarean sectionOther assisted birth Breech, researcher-reportedPerinatal asphyxia reported, researcher-reported YesNoUnknown Multiple birth, mother-reported or researcher-reported Living environment domain (mother-reported) Member of household travelled outside city, province, or country in previous 12 monthsMonthly household income (based on local area average) Household income is lower than averageHousehold income is equal to or higher than average Type of area RuralUrban Semi-ruralType of house ApartmentSeparate houseShackOther Number of bedrooms in residence 01–23 or more Number of people residing in the house 1–34–67 or more Primary source of drinking water Municipal networkPrivate wellCommunal tapsWater vendor, sachet, or bottled waterGround water Is the water treated BoiledFilteredNeither Presence of stagnant or sewage water near homeElectricity supply in household (based on reported hours or days of availability) No supplyPoor supplyIntermittent supplyRegular supply Frequency of solid waste collection, if nearby No solid waste pipe nearby, or otherWe deal with it ourselvesOnce a week or moreLess than once a week Type of toilet Pit latrine or no toiletSit or squat toilet with flush Whether house is served by wastewater networkAccess to soap (of mother) YesNoSometimes Hand washing frequency (of mother) OccasionallyFrequently Bath or shower frequency (of mother) OccasionallyFrequently

13Health-care domain (researcher reported)

14Type of ward MaternityNeonatalObstetrics and gynaecologyOther Number of beds on the ward 1–34–1011–2021 or more Bathroom on the wardMother's location on the ward First third (closest to door)Middle thirdLast third (furthest from door)Other

15Type of ward

16MaternityNeonatalObstetrics and gynaecologyOther

17Maternity

18Neonatal

19Obstetrics and gynaecology

20Other

21Number of beds on the ward

221–34–1011–2021 or more

231–3

244–10

2511–20

2621 or more

27Bathroom on the ward

28Mother's location on the ward

29First third (closest to door)Middle thirdLast third (furthest from door)Other

30First third (closest to door)

31Middle third

32Last third (furthest from door)

33Other

34Maternal domain (mother reported)

35Mother's age 20 years or younger21–35 years36 years or older Pregnancy history First pregnancyPrevious pregnancy did not result in a livebirthLive previous birth Health conditions DiabetesHypertensionImmune-compromisedMalariaTuberculosisOther infectionTyphoidOther health condition Received private health care in previous 3 monthsVisited traditional healer in previous 3 monthsHospitalised in previous 12 monthsUsed antibiotics in previous 3 monthsMother's educational status None or primarySecondary schooling or university

36Mother's age

3720 years or younger21–35 years36 years or older

3820 years or younger

3921–35 years

4036 years or older

41Pregnancy history

42First pregnancyPrevious pregnancy did not result in a livebirthLive previous birth

43First pregnancy

44Previous pregnancy did not result in a livebirth

45Live previous birth

46Health conditions

47DiabetesHypertensionImmune-compromisedMalariaTuberculosisOther infectionTyphoidOther health condition

48Diabetes

49Hypertension

50Immune-compromised

51Malaria

52Tuberculosis

53Other infection

54Typhoid

55Other health condition

56Received private health care in previous 3 months

57Visited traditional healer in previous 3 months

58Hospitalised in previous 12 months

59Used antibiotics in previous 3 months

60Mother's educational status

61None or primarySecondary schooling or university

62None or primary

63Secondary schooling or university

64Birth and neonatal domain

65Gestational age, mother-reported Term (37–41 weeks)Preterm (<37 weeks)Post-term (>41 weeks) Premature rupture of membranes, mother reported or researcher reportedDelivery type, researcher reported Natural (spontaneous vaginal delivery)Planned caesarean sectionEmergency caesarean sectionOther assisted birth Breech, researcher-reportedPerinatal asphyxia reported, researcher-reported YesNoUnknown Multiple birth, mother-reported or researcher-reported

66Gestational age, mother-reported

67Term (37–41 weeks)Preterm (<37 weeks)Post-term (>41 weeks)

68Term (37–41 weeks)

69Preterm (<37 weeks)

70Post-term (>41 weeks)

71Premature rupture of membranes, mother reported or researcher reported

72Delivery type, researcher reported

73Natural (spontaneous vaginal delivery)Planned caesarean sectionEmergency caesarean sectionOther assisted birth

74Natural (spontaneous vaginal delivery)

75Planned caesarean section

76Emergency caesarean section

77Other assisted birth

78Breech, researcher-reported

79Perinatal asphyxia reported, researcher-reported

80YesNoUnknown

81Yes

82No

83Unknown

84Multiple birth, mother-reported or researcher-reported

85Living environment domain (mother-reported)

86Member of household travelled outside city, province, or country in previous 12 monthsMonthly household income (based on local area average) Household income is lower than averageHousehold income is equal to or higher than average Type of area RuralUrban Semi-ruralType of house ApartmentSeparate houseShackOther Number of bedrooms in residence 01–23 or more Number of people residing in the house 1–34–67 or more Primary source of drinking water Municipal networkPrivate wellCommunal tapsWater vendor, sachet, or bottled waterGround water Is the water treated BoiledFilteredNeither Presence of stagnant or sewage water near homeElectricity supply in household (based on reported hours or days of availability) No supplyPoor supplyIntermittent supplyRegular supply Frequency of solid waste collection, if nearby No solid waste pipe nearby, or otherWe deal with it ourselvesOnce a week or moreLess than once a week Type of toilet Pit latrine or no toiletSit or squat toilet with flush Whether house is served by wastewater networkAccess to soap (of mother) YesNoSometimes Hand washing frequency (of mother) OccasionallyFrequently Bath or shower frequency (of mother) OccasionallyFrequently

87Member of household travelled outside city, province, or country in previous 12 months

88Monthly household income (based on local area average)

89Household income is lower than averageHousehold income is equal to or higher than average

90Household income is lower than average

91Household income is equal to or higher than average

92Type of area

93RuralUrban

94Rural

95Urban

96Semi-rural

97Type of house

98ApartmentSeparate houseShackOther

99Apartment

100Separate house

101Shack

102Other

103Number of bedrooms in residence

10401–23 or more

1050

1061–2

1073 or more

108Number of people residing in the house

1091–34–67 or more

1101–3

1114–6

1127 or more

113Primary source of drinking water

114Municipal networkPrivate wellCommunal tapsWater vendor, sachet, or bottled waterGround water

115Municipal network

116Private well

117Communal taps

118Water vendor, sachet, or bottled water

119Ground water

120Is the water treated

121BoiledFilteredNeither

122Boiled

123Filtered

124Neither

125Presence of stagnant or sewage water near home

126Electricity supply in household (based on reported hours or days of availability)

127No supplyPoor supplyIntermittent supplyRegular supply

128No supply

129Poor supply

130Intermittent supply

131Regular supply

132Frequency of solid waste collection, if nearby

133No solid waste pipe nearby, or otherWe deal with it ourselvesOnce a week or moreLess than once a week

134No solid waste pipe nearby, or other

135We deal with it ourselves

136Once a week or more

137Less than once a week

138Type of toilet

139Pit latrine or no toiletSit or squat toilet with flush

140Pit latrine or no toilet

141Sit or squat toilet with flush

142Whether house is served by wastewater network

143Access to soap (of mother)

144YesNoSometimes

145Yes

146No

147Sometimes

148Hand washing frequency (of mother)

149OccasionallyFrequently

150Occasionally

151Frequently

152Bath or shower frequency (of mother)

153OccasionallyFrequently

154Occasionally

155Frequently

156The three primary outcomes were: clinically suspected sepsis, laboratory-confirmed sepsis, and all-cause mortality during the first 60 days of life. Clinically suspected sepsis was defined as the collection of 0·5 mL or more of blood for culture from a neonate where the clinician had identified sepsis symptoms. In neonates with clinically suspected sepsis, laboratory-confirmed sepsis was defined as a positive blood culture. Blood cultures with suspected contaminants (appendix p 10) were excluded from laboratory-confirmed sepsis. All-cause mortality was defined as being reported deceased by a clinician or family member at one of the predefined follow-up points; date of death was recorded to determine neonatal age.

157BARNARDS’ countries were chosen to provide an even continental distribution between south Asia and Africa. Numerous factors including economic and health indices, populations, reported incidences of antimicrobial resistance, and neonatal sepsis, as well as veracity of data collection were considered in site selection (appendix p 3). Each site was asked to recruit all eligible mothers over a period of at least 12 months, determined by the duration of funding, and a formal sample size was not calculated.

158We estimated the incidence proportion of clinically suspected sepsis and laboratory-confirmed sepsis as the number of cases divided by the number of livebirths, multiplying by 1000 to obtain estimates per 1000 livebirths, and calculating 95% CIs with overall incidence proportions inflated for clustering of neonates within clinical sites using mixed-effects models. As clinically suspected sepsis and laboratory-confirmed sepsis outcomes were based on blood samples being taken at hospital sites, for this outcome no neonates were lost to follow-up and we could estimate incidence proportions.

159The incidence rate of all-cause mortality was calculated as the number of confirmed deaths divided by the neonate-time at risk (ie, death, 60 days of life, or final observed timepoint as a censored observation). This was multiplied by 1000 to provide a rate per 1000 neonate-days and accompanied by 95% CIs. We focused on neonate-time as our denominator since not all neonates were followed up for 60 days. The all-cause mortality rate is reported overall, by clinical site, and by sepsis status (no clinically suspected sepsis and no laboratory-confirmed sepsis, clinically suspected sepsis but no laboratory-confirmed sepsis, and clinically suspected sepsis with laboratory-confirmed sepsis [a sub-set of clinically suspected sepsis]). Overall and sepsis status-specific confidence intervals for all-cause mortality rates were inflated for clustering of neonates within clinical sites by calculating jackknife CIs.

160We report time from birth until clinically suspected sepsis and laboratory-confirmed sepsis as incidence graphs (appendix p 48), time from birth until all-cause mortality as Kaplan-Meier survival curves, and the ten most common bacterial organisms associated with laboratory-confirmed sepsis as stacked bar charts (appendix p 49).

161We fitted bivariable and multivariable regression models to investigate the associations between factors in the health-care, maternal, birth and neonatal, and living-environment domains and clinically suspected sepsis, laboratory-confirmed sepsis, and all-cause mortality. For clinically suspected sepsis and laboratory-confirmed sepsis, we fitted Poisson regression models with robust standard errors and controlling for site as a fixed effect. Findings are reported as relative risks (RRs) and 95% CIs. For all-cause mortality, we fitted flexible parametric survival models, adjusting for site and sepsis status, with the baseline hazard function modelled as a restricted cubic spline with three knots. Findings are reported as hazard ratios (HRs) and 95% CIs. For laboratory-confirmed sepsis, we fitted multivariable models adjusting for known and measured confounders based on the minimally sufficient adjustment sets implied by our directed acyclic graphs. Our analytical directed acyclic graphs (appendix pp 11–14) were informed by a review of the literature and expert opinion within the study team. We extended our Poisson regression models, with each model focusing on a single exposure of interest (preterm birth, premature breaking of waters, delivery type, and perinatal asphyxia), and adjustments made for known and measured confounders implied by the directed acyclic graphs. For the premature breaking of waters analysis, we further extended this by including a gestational age × premature breaking of waters interaction to explore whether the adjusted association between premature breaking of waters and laboratory-confirmed sepsis differed depending on gestational age. For all-cause mortality, we explored the association between sepsis status and mortality after adjusting for birth and neonatal factors and markers of socioeconomic status (including type of residence, type of toilet in home, primary source of drinking water, monthly overall household income, and electricity supply in home). All models were fitted within a multiple imputation framework (fully conditional specification with 50 imputed datasets and augmented regression used to handle perfect prediction of categorical variables). We used site, neonatal sepsis variables, and maternal age to impute missing data as we believed missing data were missing at random given observed data and these were the only complete variables that we believed would be related to the underlying missing mechanism (appendix pp 15–16). Our analysis was not adjusted for clustering of multiple births due to small numbers., All statistical analyses were conducted using Stata (v16.1).

162The funder had a minor role in the study design, including site selection and definition of primary outcome. The funder had no role in data analysis, or writing of the manuscript.

163BARNARDS enrolled 29 483 mothers and 30 557 neonates (figure 1). Recruitment took place between Nov 12, 2015, and Feb 1, 2018. Most mothers (22 678 [76·9%] of 29 483) were aged between 21 and 35 years; 33·5% (9879 of 29 483) were primigravida, 83·5% (24 621 of 29 483) reported a monthly household income lower than local average, and 66·4% (19 505 of 29 373) were educated to secondary-school level or higher (appendix pp 17–29).

164The most frequently reported sources of drinking water included municipal network (8703 [29·7%] of 29 341), private well (7883 [26·9%] of 29 341), and communal taps (6277 [21·4%] of 29 341). Most mothers (28 845 [97·9%] of 29 469) had access to soap; 59·6% reported occasional handwashing (17 563 of 29 701) and 40·1% (11 908 of 29 701) reported frequent handwashing.

165Caesarean sections were performed for 11 192 (36·7%) of 30 498 deliveries, with 8055 (72·0%) recorded as emergencies, 3084 (27·5%) reported as planned, and the remainder unknown (<1%). 1784 (42·4%) of 4205 preterm neonates, 877 (53·8%) of 1630 breech presentation neonates, 1055 (43·24%) of 2440 neonates with reported perinatal asphyxia, and 1151 (31·0%) of 3713 neonates born following PROM were all delivered by emergency caesarean section. Preterm neonates (<37 weeks) accounted for 4227 (14·6%) of 29 017 deliveries, and 1568 (5·4%) were post-term (>41 weeks). Additionally, 3715 (12·4%) of 29 936 neonates were delivered following PROM and 2440 (8·0%) of 30 550 had reported perinatal asphyxia.

166Of the 30 577 neonates enrolled, 5425 neonates presented with clinical signs of sepsis, with 1372 of these identified as laboratory-confirmed sepsis (figure 1). The overall incidence proportion of clinically suspected sepsis was 166·0 per 1000 livebirths (95% CI 97·69–234·24) and for laboratory-confirmed sepsis, 46·9 per 1000 livebirths (19·04–74·79; table 1).

167There were 877 neonatal deaths recorded within the first 60 days of life over 1 055 284 neonate-days at risk. Full follow-up data (ie, alive at or deceased by the end of the study) were available for 16 789 (54·9%) of 30 557 neonates, with the remainder censored at their final observation point (median 7 days, IQR 1–56 days). The overall incidence rate of all-cause mortality was 0·83 per 1000 neonate-days (95% CI 0·37–2·00). In neonates with no clinically suspected sepsis and no laboratory-confirmed sepsis, the incidence rate of all-cause mortality was 0·40 per 1000 neonate-days (0·25–0·62). Those with clinically suspected sepsis but no laboratory-confirmed sepsis had an all-cause mortality incidence rate of 2·86 per 1000 neonate-days (1·34–7·60) and for those with clinically suspected sepsis and laboratory-confirmed sepsis it was 5·65 per 1000 neonate-days (3·00–13·35; appendix p 30). The incidence rates of clinically suspected sepsis, laboratory-confirmed sepsis, and all-cause mortality varied across the clinical sites (Table 1, Table 2).

168Blood culture results from 31 neonates with clinically suspected sepsis contained suspected bacterial contaminants and did not meet our laboratory-confirmed sepsis definition (appendix p 10). Of the 31 neonates, five were reported deceased.

169Associations with clinically suspected sepsis are fully detailed in appendix pp 31–36. Some factors found to be associated with a higher risk of clinically suspected sepsis included private health care within 3 months before enrolment, living in a separate house compared with an apartment, having no bedrooms in the house compared with one or more rooms, drinking water from a private well, communal taps, or groundwater compared with municipal network and living in a rural environment compared with an urban environment.

170Health-care factors associated with a higher risk of laboratory-confirmed sepsis included being on a neonatal ward compared with a maternity ward (RR 7·73, 95% CI 6·31–9·46), having a bathroom on the ward compared with no bathroom (1·78, 1·28–2·46), and being situated at the entrance to or in the middle of the ward compared with furthest from the door (1·42, 1·21–1·67 for entrance and 1·42, 1·20–1·67 for middle; appendix pp 37–40).

171Maternal factors associated with a higher risk of laboratory-confirmed sepsis included maternal use of antibiotics within 3 months before enrolment (RR 1·48, 95% CI 1·22–1·82) compared with no antibiotic use, maternal hypertension compared with no hypertension (1·73, 1·47–2·04), and being hospitalised in the preceding 12 months compared with no hospitalisation (1·53, 1·27–1·85; appendix pp 37–40).

172Birth and neonatal factors associated with a higher risk of laboratory-confirmed sepsis included preterm delivery (RR 3·94, 95% CI 3·53–4·39), PROM (1·64, 1·44–1·87), caesarean section delivery compared with spontaneous vaginal delivery (1·50, 1·23–1·84 for elective and 1·92, 1·69–2·18 for emergency), reported perinatal asphyxia compared with no asphyxia (3·82, 3·38–4·31), being part of multiple births compared with singleton (1·68, 1·40–2·01), and breech delivery compared with spontaneous vaginal delivery (2·02, 1·70–2·41; appendix pp 37–40).

173Living-environment factors associated with a higher risk of laboratory-confirmed sepsis included having at least average monthly household income compared with lower than average income (RR 1·22, 95% CI 1·03–1·44), living in a household served by a wastewater network compared with those who were not (1·26, 1·11–1·43), primary source of drinking water being from a water vendor, sachet, or bottled water compared with municipal network (1·31, 1·03–1·65), and living in a rural environment, compared with urban (1·26, 1·11–1·43; appendix pp 37–40).

174The multivariable analysis identified that the associations between our exposures of interest (preterm delivery, PROM, delivery type, and reported perinatal asphyxia) and laboratory-confirmed sepsis remained statistically significant after adjusting for known confounding factors (table 3; appendix pp 11–14). There was further evidence to suggest that the association between PROM and laboratory-confirmed sepsis was stronger in term neonates (RR 1·56, 1·28–1·89) compared with preterm (0·99, 0·83–1·19) and post-term neonates (1·27, 0·55–2·93; p-value for gestational age × PROM interaction=0·004; appendix p 41).

175Neonates with either clinically suspected sepsis only or clinically suspected sepsis and laboratory-confirmed sepsis had higher rates of all-cause mortality than those without sepsis (HR 5·00, 95% CI 4·26–5·88 for clinically suspected sepsis and 8·95, 7·43–10·79 for laboratory-confirmed sepsis). Neonates situated on a neonatal ward compared with a maternity ward (1·88, 1·38–2·55) were associated with higher rates of all-cause mortality. Maternal hospitalisation in the preceding 12 months compared with no hospitalisation (1·29, 1·02–1·63) and maternal hypertension, compared with no reported hypertension (1·29, 1·05–1·59) were associated with higher rates of neonatal all-cause mortality (appendix pp 42–44).

176Preterm delivery was associated with higher rates of all-cause mortality compared with term delivery (HR 4·00, 95% CI 3·41–4·69), as was reported perinatal asphyxia compared with no reported asphyxia (3·74, 3·16–4·42) and multiple births compared with singleton (2·91, 2·42–3·51; appendix pp 42–44).

177After adjusting for birth and neonatal factors and markers of socioeconomic status, the all-cause mortality rate remained higher for those with clinically suspected sepsis (HR 2·89, 95% CI 2·42–3·44) and those with clinically suspected sepsis and laboratory-confirmed sepsis (4·39, 3·59–5·37), than compared with neonates with no sepsis within the first 60 days of life.

178Living-environment factors associated with higher rates of all-cause mortality included living in a household with a monthly income of at least average compared with lower than average (HR 1·27, 95% CI 1·01–1·60), living in a shack compared with an apartment (1·37, 1·00–1·87), and residing in a rural area compared with urban area (1·20, 1·02–1·41; appendix pp 42–44).

179Figures 2A–C are forest plots representing all statistically significant findings across clinically suspected sepsis, laboratory-confirmed sepsis, and all-cause mortality. These findings are shown in tabular format in the appendix (pp 45–47).

180Time to laboratory-confirmed sepsis by age in days at clinical diagnosis (n=1215) was examined. The greatest proportion of laboratory-confirmed sepsis cases occurred within the first 3 days of life (881 [72·5%] of 1215) and within the first 7 days of life (1078 [88·7%] of 1215; median 1 day, IQR 0–3 days). These results closely aligned with the time from birth until clinically suspected sepsis (median 1 day, IQR 0–2 days; appendix p 48).

181Of the 877 neonates who died, 66% (n=581) died within the first 7 days of life and 88% (n=770) within the first 28 days. Time from birth to all-cause mortality is illustrated in figure 3.

182In total, 49 different bacterial species causing sepsis were identified by whole genome sequencing methods. The five most frequently identified bacteria included Klebsiella pneumoniae (n=123), Klebsiella michiganensis (n=76), Staphylococcus aureus (n=63), Serratia marcescens (n=49), and Burkholderia cenocepacia (n=44). Escherichia coli (n=34), Enterobacter cloacae (n=33), and Acinetobacter spp (n=17) were identified in all countries (appendix p 49). Of 123 K pneumoniae-associated sepsis cases, 56·9% (n=70) were delivered by caesarean section, and of 34 E coli-associated sepsis cases, 58·8% (n=20) were delivered by caesarean section.

183BARNARDS assessed the incidence and factors associated with neonatal sepsis and all-cause mortality in seven LMICs representing a range of geographical and economical settings. Compared with neonates with no clinically suspected sepsis and those with no laboratory-confirmed sepsis, all-cause mortality was considerably higher among those with clinically suspected sepsis and even higher again for those with laboratory-confirmed sepsis, confirming that neonatal sepsis is a great contributor to neonatal mortality.

184The incidence of clinically suspected sepsis was highly variable across clinical sites potentially due to varying methods of clinical practice and contextual factors—eg, laboratory equipment and population served. Within the DeNIS study, incidence of clinically suspected sepsis was 143 per 1000 admissions; lower than BARNARDS. However, incidence of laboratory-confirmed sepsis was 62 per 1000 livebirths; higher than BARNARDS. Inclusion criteria for DeNIS study included admission into a neonatal intensive care unit whereas BARNARDS included all facility-born neonates. In BARNARDS, the highest incidence rate of all-cause mortality was in neonates with laboratory-confirmed sepsis (5·65 per 1000 neonate-days). Proportionately, the number of deaths recorded in those excluded due to suspected bacterial contaminants was consistent with the all-cause mortality rate in those with laboratory-confirmed sepsis.

185Across all four domains we identified associations with increased risks of clinically suspected sepsis, laboratory-confirmed sepsis, and of all-cause mortality. Poor sanitation and living conditions are associated with neonatal sepsis. We found mothers from households with no household electricity supply had an increased risk of clinically suspected sepsis and laboratory-confirmed sepsis, this could be due to vertical transmission. No evidence of an association between maternal education and neonatal sepsis was found, consistent with existing work. A counterintuitive finding was the association between clinically suspected sepsis, laboratory-confirmed sepsis, and all-cause mortality and at least average monthly household income, suggesting a higher household income is associated with a poorer neonatal outcome. This finding might reflect reporting biases, accessibility to, and availability of health care or that household income might not be representative of the wealth of the mother.

186Increased rates of neonatal sepsis have been reported in mothers with diabetes or hypertension in high-income countries. In BARNARDS, women with maternal hypertension had an increased risk of delivering a neonate with clinically suspected sepsis, laboratory-confirmed sepsis, and all-cause mortality. PROM is a recognised association with neonatal sepsis, first recorded in 1963; our findings were consistent with existing literature. PROM is associated with breech presentation and we found associations between clinically suspected sepsis, laboratory-confirmed sepsis, and breech presentation, which is unreported elsewhere; a probable cause for this finding is the known connection between PROM and neonatal sepsis. There are known links between maternal malaria and neonatal sepsis; however, these were not identified in BARNARDS. Location in the hospital after the delivery, such as being on a larger hospital ward (>11 beds), located closest to the door or in the middle of the ward, and being situated on a neonatal ward compared with a maternity ward (this would only occur if the neonate needed treatment, a probable reason for this finding) were also associated with increased risks of neonatal sepsis and all-cause mortality.

187Clinical interventions such as prenatal antibiotic use and caesarean sections were associated with neonatal sepsis. Increased risks of clinically suspected sepsis and laboratory-confirmed sepsis associated with prenatal antibiotic use might be related to the disruption of the mother's vaginal microbiota, increasing exposure to sepsis-causing bacteria—eg, Enterobacterales. Further research is needed to determine the effect of prenatal antibiotics on neonates’ health; further works could include an in-depth evaluation into antibiotic type, duration, and reason for use to provide a comprehensive understanding of this association. Caesarean sections, elective and emergency, were associated with clinically suspected sepsis and laboratory-confirmed sepsis when compared with spontaneous vaginal delivery. Emergency caesarean sections carried an increased risk, with these neonates likely to have additional risk factors. For example, 1784 (42·4%) of 4205 preterm neonates, 877 (53·8%) of 1630 breech presentation neonates, 1055 (43·2%) of 2440 neonates with reported perinatal asphyxia, and 1151 (31·0%) of 3713 neonates delivered following PROM were delivered by emergency caesarean sections. A study investigating PROM deliveries in Kosovo reported 28% caesarean sections rate. Neonates delivered via caesarean section are not exposed to the mothers’ vaginal microbiota and might be at increased risk of exposure to pathogenic bacteria, such as staphylococci and Enterobacterales. Performing caesarean sections is a reported risk factor for acquisition of extended-spectrum β-lactamase-producing Enterobacterales, bacteria which commonly cause sepsis in health-care settings. Caesarean sections inevitably lead to longer hospital stays compared with spontaneous vaginal delivery, increasing chances of neonatal sepsis. This said, our findings indicate an association between emergency caesarean sections and a lower rate of all-cause mortality, potentially because the reason for the caesarean section was unrelated to neonatal sepsis. Higher risks of clinically suspected sepsis and laboratory-confirmed sepsis were identified in women in their first pregnancy compared with those who had previously had a livebirth, congruent with existing research., A plausible explanation for this is longer labour or complications associated with first pregnancies.

188The management of neonatal sepsis in LMICs is challenging and increasingly complex due to increased levels of antimicrobial resistance and poor resources for diagnosis and treatment. Similar to other studies, we found that the majority of neonates developed sepsis within the first 24–48 h of life., This finding signals a need for improvement in rapid diagnostic methods and targeted treatment. A recent review has highlighted the use and importance of multi-omics to improve the accuracy of sepsis biomarkers and the use of health systems data for tailoring digital diagnostics. Basic interventions such as targeted infection prevention control and implementation of water, sanitation, and hygiene measures (known as WASH), which contribute to the prevention of spread of antimicrobial resistance, and therefore reduce the incidence of neonatal sepsis. Gaining an understanding of associations with neonatal sepsis and mortality is helpful for informing interventions to combat and reduce neonatal sepsis in LMICs; through BARNARDS we are able to contribute to this vital evidence base.

189BARNARDS consisted of a wide range of clinical sites, representative of usual clinical settings and some non-research sites. Our results provide an accurate characterisation of the burden of neonatal sepsis and mortality in under-served and understudied populations across several LMICs. However, we note two limitations with regard to our data and analysis. First, the definition of laboratory-confirmed sepsis is considered to be fewer than four colony-forming units of bacteria per mL of blood. Obtaining adequate volume for blood culture samples, particularly from preterm neonates, is challenging. Small volumes of blood reduce the probability of detecting bacterial organisms, which might result in underestimated prevalence of laboratory-confirmed sepsis. Second, differing interpretations of clinically suspected sepsis were apparent despite standardised operating procedures. Some laboratory-confirmed sepsis isolates were lost at clinical sites or in transit to the laboratory at Cardiff University (UK). Third, not all neonatal outcomes were recorded (lost to follow-up). Therefore, all-cause mortality might be under-reported as 60-day follow-up was challenging. The decision to analyse time to all-cause mortality, incorporating all data available and applying censoring to neonates lost to follow-up therefore yields a lower bound for these incidence estimates.

190The use of purposive sampling to select clinical sites and countries, in addition to a scarcity of screening data, means that caution should be taken when generalising these findings within countries or continents. Causal relations cannot be drawn from observational studies due to residual confounding.

191Within BARNARDS, facility-based cohort data and laboratory methods were generated to improve the identification of neonatal sepsis at a scale not previously seen within this population. BARNARDS highlights the burden of neonatal sepsis in LMICs, and identifies a multitude of factors associated with clinically suspected sepsis, laboratory-confirmed sepsis, and all-cause mortality. Several important knowledge gaps remain, including an urgent need for studies looking at simple and sustainable interventions to reduce the burden of neonatal sepsis.

192Data will be made available upon request from researchers who provide a methodologically sound proposal, following assessment from the research team, and subject to a data sharing agreement. Data will be de-identified and will strictly adhere to patient confidentiality and consent. All relevant study protocols including ethical approvals are on the BARNARDS group website (https:/www.ineosoxford.ox.ac.uk/research/barnards). Datasets specific to this study will be made available upon request following publication.

193We declare no competing interests.