Early-onset neonatal sepsis in Dhaka, Bangladesh: risk associated with maternal bacterial colonisation and chorioamnionitis

Authors


Abstract

Objective

To estimate the risk of early-onset neonatal sepsis among newborns of mothers with chorioamnionitis and/or bacterial colonisation in Dhaka.

Methods

We conducted a cohort study at a maternity centre following 600 mother–newborn pairs. Women with a positive bacterial vaginal culture or positive Group B streptococcus (GBS) rectal culture during labour were classified as colonised. Women with placental histopathology demonstrating signs of maternal or foetal inflammation were classified as having chorioamnionitis. Newborns were followed over the first 7 days of life. The primary outcome measure was physician or community health worker diagnosis of neonatal sepsis following modified World Health Organization Integrated Management of Childhood Illnesses criteria. Survival analysis was conducted with non-parametric, parametric and semiparametric models.

Results

Of the 600 mother–newborn pairs, 12.8% of newborns were diagnosed with early-onset sepsis. Five hundred and forty-three women had both colonisation and chorioamnionitis data, 55.4% of mothers were non-exposed, 31.7% were only colonised and 12.9% had chorioamnionitis regardless of colonisation status.

After adjusting for birthweight, sex, maternal characteristics and wealth, newborns of only colonised mothers developed sepsis 63% faster and had a 71% higher risk of developing sepsis than their non-exposed counterparts (RT = 0.37, 95% CI 0.14–1.03; RH = 1.71, 95% CI 1.00–2.94). Newborns of mothers with chorioamnionitis developed sepsis 74% faster and had a 111% higher risk of developing sepsis (RT = 0.26, 95% CI 0.07–0.94; RH = 2.11, 95% CI 1.06–4.21).

Conclusions

Newborns born to mothers with colonisation or chorioamnionitis developed sepsis faster and were at higher risk of developing sepsis in Dhaka.

Abstract

Objectif

Estimer le risque de septicémie néonatale précoce chez les nouveau-nés de mères atteintes d'une chorioamnionite et/ou d'une colonisation bactériennes à Dhaka.

Méthodes

Etude de cohorte dans un centre de maternité sur 600 paires mère/nouveau-né. Les femmes ayant une culture bactérienne vaginale positive ou une culture GBS rectale positive au cours du travail ont été classées comme colonisées. Les femmes avec une histopathologie placentaire démontrant des signes d'inflammation de la mère ou du fœtus ont été classées comme ayant une chorioamnionite. Les nouveau-nés ont été suivis au cours des sept premiers jours de leur vie. Le critère principal mesuré était le diagnostic de la septicémie néonatale par un médecin ou un agent de santé communautaire selon les critères modifiés de la ‘Gestion Intégrée des Maladies Infantiles de l’OMS’. L'analyse de survie a été réalisée avec des modèles non paramétriques, paramétriques et semi-paramétriques.

Résultats

Sur les 600 paires mère/nouveau-né, 12,8% des nouveau-nés ont été diagnostiqués avec une septicémie précoce. 543 femmes avaient à la fois une colonisation et des résultats de chorioamnionite. 55,4% des mères étaient non exposées, 31,7% d'entre elles étaient seulement colonisées et 12,9% avaient une chorioamnionite indépendamment de l’état de la colonisation.

Après ajustement pour le poids de naissance, le sexe, les caractéristiques maternelles et la richesse, les nouveau-nés de mères seulement colonisées ont développé des sepsis 63% plus rapidement et avaient un risque 71% plus élevé de développer une septicémie que leurs homologues non exposés (RT = 0,37; IC95%: 0,14 à 1,03; HR = 1,71; IC95%: 1,00 à 2,94). Les nouveau-nés de mères avec une chorioamnionite ont développé une septicémie 74% plus vite et avaient un risque 111% plus élevé de développer une septicémie (RT = 0,26; IC95%: 0,07 à 0,94; HR = 2,11; IC95%: 1,6 à 4,21).

Conclusions

Les nouveau-nés nés de mères avec une colonisation ou une chorioamnionite développaient une septicémie plus rapidement et étaient à risque plus élevé de développer une septicémie à Dhaka.

Abstract

Objetivo

Calcular el riesgo de aparición temprana de sepsis neonatal entre recién nacidos de madres con corioaminionitis y/o colonización bacteriana en Dhaka.

Métodos

Estudio de cohortes en una maternidad en el que se siguió a 600 parejas madre-neonato. Se clasificaron como colonizadas aquellas mujeres con un cultivo vaginal positivo para bacterias o un cultivo rectal positivo para estreptococo del grupo B durante el parto. Las mujeres con una histopatología de la placenta con signos de inflamación materna o fetal fueron clasificadas como padeciendo corioamnionitis. Los neonatos se siguieron durante sus primeros siete días de vida. El parámetro de interés primario era el diagnóstico del médico o trabajador sanitario comunitario sobre la sepsis neonatal, siguiendo los criterios de la OMS para la Atención Integrada a las Enfermedades Prevalentes en la Infancia (AIEPI). Se realizó un análisis de supervivencia con modelos no paramétricos, paramétricos y semiparamétricos.

Resultados

A un 12.8% de los neonatos, pertenecientes a las 600 parejas madre-neonato, se les diagnosticó sepsis de aparición temprana. 543 mujeres tenían datos tanto de colonización como de corioamnionitis. Un 55.4% de las mujeres no estaban expuestas, 31.7% estaban solo colonizadas y 12.9% tenían corioamnionitis, independientemente del estatus de colonización. Después de ajustar para peso al nacer, el sexo, las características maternas y el nivel de riqueza, los neonatos de madres solo colonizadas desarrollaban sepsis un 63% más rápido y tenían un 71% de mayor riesgo de desarrollar sepsis que sus contrapartes no expuestas (RT = 0.37, 95% CI 0.14–1.03; RH = 1.71, 95% CI 1.00–2.94). Los neonatos de madres con corioamnionitis desarrollaban sepsis un 74% más rápido y tenían un riesgo un 111% mayor de desarrollar sepsis (RT = 0.26, 95% CI 0.07–0.94; RH = 2.11, 95% CI 1.06–4.21).

Conclusiones

Los neonatos nacidos en Dhaka de madres con colonización o corioamnionitis desarrollaron sepsis más rápido y tenían un mayor riesgo de desarrollar sepsis.

Introduction

Neonatal infections account for an estimated 718 000 deaths per year or approximately 23.4% of the world's 3.1 million annual neonatal deaths in 2010 (Liu et al. 2012). Newborns are most vulnerable during the first week of life. As many as 42% of deaths in the first week of life may occur from infections, often defined as sepsis, but including pneumonia, bacteremia and meningitis (Thaver & Zaidi 2009). The causes of neonatal infection, the risk factors associated with them and their modes of transmission remain poorly understood in developing countries, delaying the development of effective interventions to prevent and treat neonatal infections (Bang et al. 2001).

In developed countries, chorioamnionitis or vaginal Group B streptococcus (GBS) colonisation at the time of delivery are strongly associated with early-onset neonatal GBS sepsis (OR 6.42, 95% CI 2.32–17.80; OR 204, 95% CI 100–419 respectively) (Benitz et al. 1999). Group B streptococcal sepsis is uncommon in South Asia and not much known about the prevalence of maternal vaginal–rectal colonisation or chorioamnionitis in this setting (Luck et al. 2003). Because maternal vaginal–rectal floral patterns in South Asia differ from those that predominate in the United States, we are not able to determine interventions based on the US example. Nevertheless, pathogens found in early-onset infection in South Asia, such as Klebsiella species and Escherichia coli, are common colonisers of the enteric tract and may play a role in ascending infections (Watt et al. 2003; Zaidi et al. 2004).

We hypothesise that newborns born to women with maternal colonisation or chorioamnionitis will have an increased risk of developing early-onset neonatal sepsis compared to newborns born to women without colonisation or chorioamnionitis in Dhaka, Bangladesh.

Methods

Study design

We conducted a cohort study, maternal origins of neonatal infection (MONI), at a maternity centre in Dhaka, Bangladesh following 600 mother–newborn pairs from 15 January 2011 to 31 October 2011. The maternity centre was operated by our partner non-governmental organisation, Shimantik. Pregnant women were enrolled during antenatal care visits or upon presentation to the maternity centre in labour. Newborns born to these women via vaginal deliveries were followed from birth through the first 7 days of life. We included women who were 30 gestational weeks or more and planned to deliver at the maternity centre. Women, who presented with foetal distress, obstructed labour, haemorrhage or severe pre-eclampsia were excluded to facilitate their need for urgent care. Women who used antibiotics or steroids within 2 weeks of labour were also excluded from the study. Newborns were excluded if they were born with birth injuries or surgical conditions.

Shimantik recruited four paramedics and five community health workers for primary data collection. Paramedics completed higher secondary school (12 years) and the national paramedics course. Community health workers completed at least secondary school (10 years) education. Two medical officers were part of the study team. One medical officer supervised field activities, and the other officer monitored quality assurance. Staff received 2 weeks of intensive training by a paediatrician and local medical officer using the World Health Organization (WHO) Caring for the Newborn at Home training course for community health workers (UNICEF & World Health Organization 2009). Sessions included presentations on basic principles, exercises and role plays to recognise the clinical signs and symptoms of sepsis and field experiences in patient homes. Written exams and standardised observations of staff skills were done to assess competency.

Maternal colonisation and chorioamnionitis exposure

Study paramedics interviewed enrolled women to collect demographic data on: maternal age, maternal education, antenatal care provider type and receipt of tetanus toxoid as a proxy for access to health care.

At least one study paramedic was present 24 h 7 days a week in the labour and delivery room to assess maternal risk factors including stage and duration of labour, rupture of membranes, intrapartum temperature, number of vaginal exams performed, amniotic fluid colour, hand washing by health workers and maternal reproductive tract colonisation status.

During labour, study paramedics collected vaginal swabs for Escherichia coli, Klebsiella pneumonia, Pseudomonas species, Acinetobacter species, GBS, non-group B streptococcus, Staphylococcus aureus or Staphylococcus species other than S. aureus cultures and rectal swabs for GBS testing. Swabs were transported daily to the Department of Microbiology at Dhaka Shishu Hospital using an Amies transport medium and processed within 24 h of collection. Women with a positive bacterial vaginal culture or positive GBS rectal culture were classified as colonised. Culture results were not available prior to delivery.

Paramedics also collected data on neonatal characteristics such as sex, birthweight to the nearest 100 g and gestational age based on ultrasound report or maternal report of last menstrual period.

Four sections of the placenta were collected immediately after delivery: a section from the extraplacental membranes, a cross-section of the umbilical cord and two sections of the central placental disc (one from the foetal surface and one from the maternal surface). Sections were fixed in 10% buffered formalin and then embedded in paraffin until processing. Histopathology slides were blindly reviewed by a pathologist, and readings were verified by an expert placental histopathologist at the conclusion of the study. Classification of histological chorioamnionitis was based on either maternal neutrophilic inflammation in the chorionic plate or transmigration of foetal neutrophils through the umbilical vessels and the chorionic plate.

Early-onset neonatal sepsis outcome

The primary outcome measure was early-onset neonatal sepsis defined as a positive blood culture or clinical signs of early-onset neonatal sepsis identified by a physician or community health worker in the first 7 days of life. Clinical signs of sepsis were defined as the presence of convulsion, severe chest indrawing, fever >37.5 °C, hypothermia <35.5 °C, movement only with simulation, history of poor feeding, or fast breathing >60 breaths per minute confirmed twice, following the WHO Young Infant Integrated Management of Childhood Illnesses algorithm (World Health Organization & UNICEF 2008). In our analyses, we excluded the fast breathing criteria due to low specificity of this sign. Classification of the outcome status was determined by staff unaware of the maternal colonisation and chorioamnionitis status.

A physician examined each newborn in the maternity centre within 12 h of birth. Trained community health workers conducted newborn home visits on the third and seventh days of life. During these visits, community health workers collected data on maternal report of neonatal illnesses, examined newborns for clinical signs or symptoms of sepsis and collected data on health-seeking behaviours. On days of life 2, 4, 5 and 6, community health workers conducted phone follow-ups collecting the same information as that obtained during the home visits with the exception of a physical exam.

Community health workers identified sick newborns based on a checklist of clinical signs or symptoms collected from maternal report and/or the community health worker's physical examination of the newborn. Newborns identified as sick by community health workers or by maternal report were evaluated by a study physician. A kappa statistic was used to evaluate agreement between the community health workers and physicians' assessments.

We collected peripheral blood cultures from newborns with clinical signs of sepsis. However, there were low rates of compliance with referrals to the tertiary care centre for blood cultures. We also collected cord blood gases and cord blood cultures on a sample of newborns to improve our precision in diagnosing sepsis. Newborns with clinical signs of sepsis but with asphyxia were excluded from the analyses as a case of sepsis. We defined asphyxia as cord blood pHs <7.0 and partial pressures of carbon dioxide >70 mmHg. For cases without cord blood gases, we used physician diagnosis of asphyxia based on clinical signs: no cry more than 5 min after birth, resuscitation with bag and mask, negative cord or peripheral blood cultures if available or normal white blood cell counts and differential. We also conducted a sensitivity analysis classifying those with asphyxia as having sepsis.

This study received ethical approval from the Johns Hopkins Bloomberg School of Public Health Committee on Human Research and the International Centre for Diarrheal Disease Research, Bangladesh Ethical Review Committee. All study participants provided informed consent.

Statistical methods

In our analysis, the event of interest was early-onset neonatal sepsis within 7 days of birth, and the exposure was the presence of maternal colonisation or chorioamnionitis. Women were divided into three groups: non-exposed, colonisation only and chorioamnionitis with or without colonisation (hereafter referred to as chorioamnionitis). Women in the colonisation or chorioamnionitis groups were referred to as exposed. We used a chi-square test to compare the difference in the proportion of early-onset neonatal infection among newborns of exposed mothers compared to newborns of non-exposed mothers.

To model the risk of developing early-onset neonatal sepsis over time, we used non-parametric models (Kaplan–Meier), parametric models based on the Weibull distribution (Cox et al. 2007) and semiparametric (classical proportional hazards) models. Our origin was time of birth. The timescale was number of days since birth. Censoring occurred at diagnosis of asphyxia, death, loss to follow-up or at the end of the follow-up period (7 days).

To obtain the estimates of the three (for non-exposed, colonised only, and chorioamnionitis with or without colonisation groups) hazard functions and their ratios, we fit the data using a Weibull model defined by locations β and scale σ whereby the pth percentile is exp{β + σ log [−log(1−p)]}. The conventional analysis corresponds to allowing for the three groups to have different locations [i.e. non-exposed approximately Weibull(β0, σ); colonised approximately Weibull(β1, σ); chorioamnionitis approximately Weibull(β2, σ)]. In this setting, the relative hazards (RH) of the colonised and chorioamnionitis relative to the non-exposed are constant and equal to exp[(β0−β1)/σ] and exp[(β0−β2)/σ], respectively. In parallel, the relative times (RT) of the colonised and chorioamnionitis to non-exposed are also constant and equal to exp(β1−β0) and exp(β2−β0) respectively. To address for non-proportional hazards and times, we also allowed the sigmas to vary in the Weibull model.

A likelihood ratio test with two degrees of freedom (6 parameters allowing for heterogeneous sigmas – minus 4 parameters allowing for a common sigma) was carried out to test for non-proportionality. The Weibull model with a common sigma agrees with the classical semiparametric proportional hazards regression approach; but it in addition provides estimates of the hazard function parameters and provides elements to decide the nature of the hazards of sepsis over time. An attractive feature of the parametric approach is the ability to consider the events recorded on day t as actually happening sometime between − 0.5 and + 0.5 (i.e. interval censoring).

Using multivariable models, we investigated the association between early-onset neonatal infection and maternal exposures adjusting for the potential confounding effects of low birthweight, sex, maternal education, maternal age, type of antenatal care provider, number of vaginal exams and wealth quintiles. Wealth quintiles were created using principal components analysis conducted with the following variables: type of housing materials for the roof, wall and floor; source of water supply; type of latrine; household number; household number under 5 years; and number of rooms where household members sleep (Vyas & Kumaranayake 2006).

Analyses were performed using STATA v12 (StataCorp, College Station, TX), and graphs of the hazards were produced in R.

Results

Of the 600 mother–newborn pairs, seventy-seven newborns (12.8%) were classified with early-onset neonatal sepsis. Of these 77 newborns, ten were hospitalised and received antibiotics. Eight of the hospitalised newborns improved and were discharged; two died. Two additional newborns with sepsis died at home. The majority of newborns with sepsis (77.9%) were diagnosed in the first 3 days of life: 33 on the first day of life, 11 on the second day and 16 on the third day. There were slightly more newborns with difficulty feeding among those who were diagnosed with sepsis later compared to those who were diagnosed with sepsis in the first 3 days of life (18.6% vs. 4.5%, P = 0.09) although this was not statistically significant. For the other clinical signs, there were no differences in the proportion of newborns diagnosed with sepsis in the first 3 days of life compared to newborns diagnosed with sepsis later.

Twenty-eight (4.7%) newborns were diagnosed as having asphyxia. Seventeen of the 600 newborns (2.8%) were lost to follow-up. Physicians diagnosed or confirmed sepsis in 44 newborns. Kappa statistics show substantial agreement (к = 0.63) between assessments of very severe disease by community health workers and physicians. Only twelve peripheral blood cultures were obtained among those who were diagnosed with early-onset neonatal sepsis all of which were negative. One cord blood culture of the 209 cord blood cultures was positive (Escherichia coli); however, the newborn had no clinical signs of sepsis.

Two-hundred and thirteen (36.4%) women were colonised of the 585 women with maternal colonisation data; 185 singly colonised and 28 cocolonised. Among women singly colonised, the most common organisms were S. aureus (7.4%), non-GBS streptococcus (6.8%) and GBS (6.2%). Five hundred fifty-two mothers had placental histopathology data. Of them, 70 (12.7%) had chorioamnionitis. A total of 543 women had both colonisation and chorioamnionitis data of which 301 (55.4%) mothers were non-exposed, 172 (31.7%) were only colonised, and 70 (12.9%) had chorioamnionitis with or without colonisation. Of the 70 women with chorioamnionitis, 19 (27.1%) were also colonised.

Demographically, the three groups (non-exposed, only colonised, chorioamnionitis) were similar in maternal age, maternal education, antenatal care visits by provider type, receipt of tetanus toxoid and wealth quintiles. Male newborns were more likely to be from the colonised or chorioamnionitis group (P = 0.05). Newborns in the colonised or chorioamnionitis group were more likely to be premature (P = 0.12); however, this was not statistically significant. There was little difference in birthweight across the three groups. Women with chorioamnionitis were more likely to have a maternal temperature >99°F (P = 0.06). Across the three groups, there was no difference in the proportion of women with premature rupture of membranes, prolonged rupture of membranes, and more than two vaginal exams (Table 1).

Table 1. Demographic data, neonatal baseline characteristics, risk factors by exposure group (non-exposed, colonized only, or chorioamnionitis with or without colonization)
 Non-exposed n = 301 (%)Colonized only n = 172 (%)Chorio ± colonization n = 70 (%)P-valuea
  1. a

    P-values without missing data.

Demographic data
Maternal age
≤22 years148 (49.2)85 (49.4)37 (52.9)0.85
>22 years153 (50.8)87 (50.6)33 (47.1)
Maternal schooling received
No58 (19.3)32 (18.6)15 (21.4)0.88
Yes243 (80.7)140 (81.4)55 (78.6)
ANC visit
By a non-physician213 (70.8)128 (74.4)58 (82.9)0.08
By a physician77 (25.6)40 (23.3)9 (12.9)
Missing11 (3.7)4 (2.3)3 (4.3)
Tetanus toxoid during ANC
No29 (9.6)16 (9.3)12 (17.1)0.15
Yes272 (90.4)156 (90.7)58 (82.9)
Wealth
Quintile 1 (richest 20%)56 (18.6)36 (20.9)13 (18.6)0.34
Quintile 257 (18.9)33 (19.2)18 (25.7)
Quintile 364 (21.3)27 (15.7)16 (22.9)
Quintile 459 (19.6)26 (15.1)9 (12.9)
Quintile 565 (21.6)50 (29.1)14 (20.0)
Neonatal characteristics
Sex
Female147 (48.8)66 (38.4)26 (37.1)0.05
Male133 (44.2)91 (52.9)39 (55.7)
Missing21 (7.0)15 (8.7)5 (7.1)
Gestational age
Preterm < 37 weeks24 (8.0)21 (12.2)10 (14.3)0.12
Full-term ≥ 37 weeks232 (77.1)120 (69.8)48 (68.6)
Missing45 (15.0)31 (18)12 (17.1)
Birth weight
Low birth weight < 2500 g54 (17.9)27 (15.7)13 (18.6)0.81
Birth weight ≥ 2500 g244 (81.1)141 (82.0)55 (78.6)
Missing3 (1.0)4 (2.3)2 (2.9)
Maternal risk factors/clinical signs
Premature rupture of membranes (PROM)
Yes25 (8.3)21 (12.2)4 (5.7)0.21
No276 (91.7)151 (87.8)66 (94.3)
Prolonged rupture of membranes
≥8 h126 (41.9)73 (42.4)24 (34.3)0.46
<8 h175 (58.1)99 (57.6)46 (65.7)
Number of vaginal exams
≥3155 (51.5)84 (48.8)40 (57.1)0.44
<3146 (48.5)88 (51.2)29 (41.4)
Missing001 (1.4)
Maternal temperature
>99°F10 (3.3)8 (4.7)7 (10.0)0.06
≤99°F284 (94.4)160 (93.0)62 (88.6)
Missing7 (2.3)4 (2.3)1 (1.4)

There was a greater proportion of newborns with sepsis among those of colonised only mothers or mothers with chorioamnionitis (41/242, 16.9%), compared to those of non-exposed mothers (31/301, 10.3%) (χ2 = 5.15, P = 0.023). The cumulative incidence of sepsis at 7 days of life was 10.7% among newborns of non-exposed mothers, 17.3% among newborns of only colonised mothers and 18.9% among newborns of mothers with chorioamnionitis (Figure 1).

Figure 1.

Kaplan–Meier cumulative incidence of early-onset neonatal sepsis by exposure status.

Newborns of exposed women developed sepsis faster and had a higher risk of developing sepsis than their counterparts who were born to non-exposed women. Newborns of colonised only women developed sepsis 59% faster than newborns of non-exposed women (RT = 0.41, 95% CI 0.16–1.06). Newborns of mothers with chorioamnionitis developed sepsis 69% faster than newborns of non-exposed women (RT = 0.31, 95% CI 0.09–1.03). Similarly, the risk of sepsis was higher among newborns of colonised only women (RH = 1.64, 95% CI 0.98–2.72) and even higher among newborns of women with chorioamnionitis (RH = 1.91, 95% CI 1.00–3.65) than the risk of sepsis among newborns of non-exposed women (Figure 2).

Figure 2.

Hazard rates of sepsis by exposure status, conventional Weibull model.

We adjusted for birthweight, sex, maternal education, maternal age, antenatal care provider, number of vaginal exams done and wealth quintiles. After adjusting for these potential confounders, maternal chorioamnionitis was the strongest risk factor of neonatal sepsis with a relative time of 0.26 (95% CI 0.07–0.94) and hazard ratio of 2.11 (95% CI 1.06–4.21) both of which are statistically significant, followed by maternal colonisation with a relative time of 0.37 (95% CI 0.14–1.03) and hazard ratio of 1.71 (95% CI 1.00–2.94). Sensitivity analysis comparing whether there would be a difference in hazard ratio when classifying asphyxia cases as sepsis found a similar but smaller magnitude of the effect in the same direction as the primary analysis.

Discussion

Maternal bacterial colonisation and chorioamnionitis were positively associated with early-onset neonatal sepsis. Newborns born to mothers with either colonisation or chorioamnionitis developed sepsis faster and were at higher risk of developing sepsis during the first 7 days of life. While controlling for potential confounders, maternal chorioamnionitis remained the strongest risk factor for neonatal sepsis, suggesting it plays a large role in the development of sepsis. When also adjusting for gestational age, we saw a similar trend which was not statistically significant due to the amount of missing data on ultrasound dates and last menstrual period dates.

Although exposure to colonisation or chorioamnionitis accounts for an additional 6.6% or 8.2% of early-onset neonatal sepsis cases in this study, the cumulative incidence of neonatal sepsis among non-colonised women was still high (10.7%), suggesting there are other possibly environmental factors such as unhygienic delivery, feeding practices or person-to-person transmission that could account for this high rate. Our estimated incidence of early-onset neonatal sepsis of 12.8% is similar to findings in a larger community-based study of early-onset neonatal sepsis defined by the WHO Young Infants criteria (World Health Organization & UNICEF 2008) for very severe disease (13.4%) (Mitra 2012).

This study has several limitations. The majority of women deliver at home in Bangladesh, whereas our study occurred at a first-level facility. Women who deliver at facilities may be inherently different from those who deliver at home, having failed trial of labour at home, a higher risk of complications or higher socioeconomic status. These results are generalisable to women who deliver at first-level facilities.

Accurate diagnosis of sepsis is difficult because the signs and symptoms of neonatal sepsis are non-specific. Pathogen-specific culture data would have provided stronger evidence for an association between colonisation and early-onset neonatal sepsis. However, while blood cultures are specific definitive tests for sepsis, many newborns with sepsis may not have any positive cultures (Gerdes 1991, 2004; Benitz 2010). In Bangladesh, obtaining laboratory-confirmed diagnoses of sepsis is challenging because of the lack of resources and poor referral compliance to higher-level facilities (Darmstadt et al. 2010).

To increase the specificity of our outcome, we excluded fast breathing from our case definition and those classified with other diseases that have a similar symptomatology as the WHO Young Infants algorithm. As newborns of exposed and non-exposed mothers were non-differentially classified as having sepsis, such non-differential misclassification would only underestimate our effect and bias our results towards the null.

This is one of the first studies in South Asia to estimate the risk of developing early-onset neonatal sepsis among newborns of mothers with colonisation or chorioamnionitis. One study in India, limited by a small sample size, examined maternal colonisation and early-onset neonatal infection but did not find an association between maternal vaginal colonisation and neonatal clinical sepsis (Muthusami et al. 2007). These findings have implications for future studies and policies. Pathogen-specific studies on colonisation, chorioamnionitis and early-onset neonatal sepsis would allow us to target disease pathways and focus antibiotic choices. A community-based study of the relationship between maternal colonisation, chorioamnionitis status and early-onset neonatal sepsis would yield more generalisable findings for Bangladesh.

Given the close relationship between the mother and newborn, we could focus on preventing early-onset neonatal sepsis by testing interventions targeting the mother. We know intrapartum antibiotic prophylaxis decreased the rates of early-onset GBS sepsis in the United States (Schrag et al. 2000). In settings like Bangladesh, interventions like intrapartum antibiotic prophylaxis could prevent early-onset neonatal infection from chorioamnionitis and from the common vaginal colonisers such as Staphylococcus aureus, non-group B streptococcus or Escherichia coli in addition to GBS. In Panama, a non-randomised, non-blinded trial of a single dose of a broad-spectrum antibiotic, ceftriaxone, to high-risk women (women with premature onset of labour before 37 weeks of gestation and prolonged rupture of membranes greater than 12 h) decreased the rate of sepsis-like illness in the first 5 days of life (8.1% vs. 3.1%; P = 0.06) (Saez-Llorens et al. 1995). Although our study focused on maternally acquired neonatal sepsis, a comprehensive neonatal survival package would also include interventions for sick newborns.

Conclusion

Maternal colonisation and chorioamnionitis are risk factors for neonatal sepsis in Dhaka. Maternal interventions to prevent or treat these infections should be developed in these settings.

Acknowledgements

We thank Marzia Zaman, Taufiqur Rahman, Qazi Sadequr Rahman and Abu Salaheen from the International Centre for Diarrheal Disease Research, Bangladesh for their time in the field and for data management. We are indebted to Kazi Moksedur Rahman and the field team of paramedics and community health workers from Shimantik who worked tirelessly collecting data. We are grateful to Maksuda Islam, Anik Sarkar and Iftekhar Quasem from the Child Health Research Foundation, Bangladesh for their laboratory contributions and management of sick study newborns. We thank Mukarram Ali and Nadira Majid, pathologists at Delta Hospital Limited, for processing the placental slides. Thank you to Alvaro Muñoz at Johns Hopkins Bloomberg School of Public Health for his statistical expertise and to Daniel Roth at Hospital for Sick Children, Toronto, for his field expertise. We would like to thank the mothers and newborns who participated in the study and gave their time. This research was supported by the Thrasher Research Fund Early Career Award and a grant from the National Center for Research Resources, a component of the National Institutes of Health (NIH) and the NIH Roadmap for Medical Research.

Ancillary