Incidence and risk factors of neonatal bacterial infections: a community-based cohort from Madagascar (2018–2021)

Study design

The BIRDY project was a prospective multicentric community-based mother and child cohort. The BIRDY study was initially conducted between 2012 and 2018 in Madagascar, Senegal and Cambodia in both urban and rural areas (BIRDY 1). The project continued in Madagascar only between September 2018 and September 2021. The methodology of the BIRDY project has already been described in detail elsewhere [5] and the neonatal follow-up was similar between BIRDY 1 and BIRDY 2.

Study area and recruitment

Study population included all neonates born in 3 neighborhoods (Avaradoha, Besarety, and Soavinadriana) of Antananarivo (the capital of Madagascar, with a catchment area population of 14 997) and those of the rural city of Moramanga (catchment area population of 17 159). All pregnant women in the study areas were identified with help of the community healthcare workers in order to exhaustively include all live births at delivery. This methodology allowed us to enroll neonates born at home, who would have been missed in case of neonates recruitment directly at delivery in health-care facilities. Inclusion criteria for pregnant women included usual residence in the study area with no intention of moving during the follow-up period, no objection to the ongoing research or to the collection of biological samples and having given written informed consent. Whereas in BIRDY 1 pregnant women were included during the third trimester of pregnancy, they were included from the first trimester of pregnancy in BIRDY 2.

Once enrolled, one antenatal visit was planned monthly and pregnant women were regularly contacted around the due date, so that the investigation team could be present as close as possible to the delivery (within 24 h). At delivery, neonates inclusion criteria were live newborns born to parent living in the study area with no intention of moving during the follow-up period, and whose parents were informed and consented to the ongoing research. To ensure the exhaustiveness of live-birth recruitment, all newborns meeting the inclusion criteria could be included directly at birth at the health care facility, even if their mothers had not been enrolled during their pregnancy.

Newborns follow-up: community based surveillance

At birth, newborns presenting a risk factor for infection, including premature rupture of membrane (> 24 h), fetid amniotic fluid, dystocic delivery, antibiotic therapy or maternal fever at delivery, were examined by a physician. Then, during the first three months of life, newborns were actively and passively monitored. Home visits made by study investigator were planned as part of the active monitoring: two visits were scheduled during the first week of life, then one visit per week until the end of the first month, and then two visits per month until the third month. Active follow-up ended-up at 3 months. These routine check-ups were conducted in order to minimize number of missed suspected infection and to obtain anthropometric measurements. Passive monitoring consisted of surveillance of the infant by the mother, who was asked to contact a study investigator if the infant showed signs of infection. For this purpose, mothers were taught the main signs of infection and how to take the infant’s temperature. After 3 months, 3 visits were planned at 6, 9 and 12 months to evaluate child’s growth and neurocognitive development.

Assessment and management

During follow-up, in presence of infection criteria, neonates were referred to a study collaborating physician or directly at the referring hospital in presence of signs of severity. If severe infection was suspected, blood tests and culture were systematically drawn and others samples (such as urine analysis, stool culture, lumbar puncture) performed as defined by an algorithm which was based on the WHO recommendations [6]. Antimicrobial drugs were empirically administered according to the WHO and at discretion of the physician. Bacteriological samples were taken as much as the possible before first antibiotic administration. Bacteriological results were reported to the physicians. Decisions regarding patient care and antimicrobial drug treatments were left to the physicians to decide according to local protocols.

Sampling and bacteriologic analyses

Biological samples were taken by healthcare team who collaborated with the project and who took care of the child in case of illness. If a severe bacterial infection was suspected, a venous blood culture was taken, with the volume of blood collected based on child’s weight. Amount collected varied from 0.5 ml for children under 1 kg to 2 ml for those weighing up to 6 kg. In cases of diarrhea, a sterile container was used to collect a stool sample equivalent to one tablespoon in quantity, ensuring the integrity of the sample. All collected samples (including blood, stool, urine, and cerebrospinal fluid) were immediately packaged and transported in a secure, cooled container to the microbiological laboratory at the Institut Pasteur in Madagascar to ensure rapid delivery.

Blood samples were incubated using the BACTEC automated system for up to 5 days at a temperature of 35 ± 2 °C. If positive, a direct microscopic examination of broth and gram staining were performed. Fresh blood agar and chocolate media were used for bacterial isolation, and bacterial strains were identified using matrix-assisted laser desorption/ionization time-of-flight (MALDI-TOF) mass spectrometry. Antimicrobial susceptibility testing was done by disk diffusion. The methods recommended by the French Society for Microbiology (FSM) were used for bacterial isolation, identification, and the corresponding breakpoint guidelines were followed to interpret results of antimicrobial susceptibility testing. All Enterobacterales that were resistant to third-generation cephalosporin underwent a double-disk synergy test to detect ESBL producers. This test was performed systematically and followed guidelines established by FSM. Escherichia coli ATCC 25,922 was used for quality control strains.

Urine samples were examined macroscopically and microscopically, followed by leukocyte count and Gram staining. Cultures were inoculated on selective media and incubated for 24–48 h at 35 ± 2 °C. Isolated colonies were identified with MALDI-TOF. CSF samples were visually classified and counted for red and white blood cells. Quantitative cytology was performed after cytocentrifugation and gram staining. Specific pathogens were targeted for soluble antigen detection. Two drops of CSF were inoculated on agar and incubated for 5 days. Isolated colonies were identified with MALDI-TOF. All susceptibility testing were done according to the FSM guidelines.

Data collection

Study investigators were responsible for collecting epidemiological data. During pregnancy, study investigators collected socioeconomic data and maternal information (medical and gynaecological history, treatment, obstetrical follow-up). Information on labour and delivery were also collected, including risk factors for infection, as well as data on first neonatal assessment: vitality score (APGAR score), anthropometric data (weight, height, head circumference), neonatal resuscitation, if any.

At every home visit, study investigators recorded anthropometric data (weight, height, cephalic and brachial circumference) and presence of any signs of severe infection in the infant. Medical assessments carried out during follow-up, if any, were also collected, including symptoms and clinical signs, treatments carried out, and any diagnosis retained. Finally, data on any deaths or lost to follow-up were recorded.

The end of the BIRDY 2 study period (March to August 2020) overlapped with Covid-19 epidemic in Madagascar (first case of Covid 19 detected in March 2020 in Madagascar), while 51 neonates were born and followed during this period. In Madagascar, restrictive measures (including national lockdown), which were adapted according to the evolution of the situation, were implemented between March to October 2020. Internal procedures have been put in place to ensure safety of the mothers, their newborns and the investigation team while adapting the research protocol to reach the scientific objectives of the project: active phone monitoring by investigation team for example.

Sample size

Previous findings in the same study area revealed an incidence of confirmed neonatal infections of 15.2 [10.6–21.8] per 1,000 live births [5]. Based on these prior findings, a sample size of 600 newborns would provide an estimate of the incidence of neonatal infections with a precision of approximately ± 2%, assuming a 95% confidence level. This sample size is appropriate for estimating the incidence of neonatal infections in the target population.

Study definition

We defined possible Severe Bacterial Infection (pSBI) episode as presence of at least one sign among the following: chest indrawing, tachypnea (> 60/min), hypo or hyperthermia (axillary temperature  37.5 °C, respectively), lethargy, convulsion or poor feeding. Neonates presenting with tachypnea as the only sign were not considered as cases because of low specificity of this sign when isolated [7]. All cases were reviewed by a committee composed of a medical epidemiologist, a neonatologist and an infectious disease specialist, to classify them and exclude non-severe cases and differential diagnoses. For example, post-vaccination fever, viral bronchiolitis and gastroenteritis, conjunctivitis and localised omphalitis were not considered as pSBI.

We defined confirmed Severe Bacterial Infection (cSBI) episode as the presence of clinical signs of severe infection (as defined above) and a positive culture from normally sterile sites: blood or cerebrospinal fluid or urine. Each case was classified as a cSBI or as not clinically relevant by the same expert committee. In particular, cultures positive for bacteria suggestive of contamination such as coagulase negative staphylococcus isolated in blood culture in a clinical context not suggestive of infection, were not considered as a confirmed SBI.

The neonatal follow-up period was defined from birth to 90 days of life. Early-onset sepsis were defined as those occurring in the first week of life, and late-onset sepsis as those occurring from 7 to 90 days of life [8].

In order to compare our findings with literature with follow-up periods varying from 1 to 3 months, we also considered 1-month (30 days) and 2-month period (60 days) of follow-up.

Multi-drug-resistance (MDR) was defined as acquired non-susceptibility to at least one agent in three or more antimicrobial categories [9].

We defined omphalitis as an umbilical discharge or redness extending to umbilical base, regardless of extent or severity.

Statistical analysis

We used descriptive statistics (proportions, mean, and SDs) to summarize characteristic of mothers and neonates. We compared difference in proportions and means by using the X₂ and Student-t-test, respectively. If the validity conditions for those parametric tests were not met, we used non-parametric tests: Fisher’s exact test for categorical variables and Mann-Whitney test for quantitative variables. We estimated the incidence of severe neonatal bacterial infection per 1,000 live births. The incidence rates with 95% confidence interval (95% CI) were calculated according to the criteria of possible and confirmed SBI. We calculated incidence rates of pSBI and cSBI for 30 days 60 days and 90 days of neonatal follow-up. The comparison of the different incidence rates were done using “epiR” package in R. Time to the first pSBI and cSBI were described using Kaplan-Meier survival curves and compared using the log-rank test.

We aimed to identify if low birth weight and factors related to delivery (exposures of primary interest: maternal fever, fetid amniotic fluid, premature rupture of the membranes) were associated to early onset pSBI, as early onset infections represented the majority of the overall burden of neonatal severe infection. We performed a Cox proportional hazards regression model and association were expressed as hazard ratios (HRs) with 95% CI. The hypothesis of proportionality of risks was tested for each variable in the model, using the Schoenfeld residual test. Variables that did not respect the assumption of proportional hazards were defined as stratification variables in the multivariable model. We first performed univariate analysis and, factors associated with the outcome with a p-value less than 0.20 in univariate analysis were entered in a multivariable Cox model. A manual stepwise backward procedure was carried out to identify factors independently associated with the outcome. Site (semi-rural or urban) and sex of newborn were forced in the final multivariate model.

Missing data were rare (< 1%), except for the variable “fetid amniotic fluid” with 26 missing data (5%). Newborns with missing data were excluded from the multivariate Cox model and we conducted complete case analyses [10, 11]. Then, we also performed sensitivity analyses, by considering that all missing data for the variable “fetid amniotic fluid”, corresponded firstly to non-fetid amniotic fluid and secondly to the opposite situation.

Some variables were correlated, such as difficult delivery and need for resuscitation at birth or such as primigravidae and maternal age. We included in the multivariate Cox model need for resuscitation at birth because this variable was the most strongly associated with neonatal infection; and maternal age because primigravidae does not respect the assumption of proportional hazards.

In a second step, to document strength and direction between the outcome and variables that did not respect the proportional hazards assumption, a parametric model was carried out with an accelerated failure time (AFT) approach, using a log-normal distribution. The overall analysis strategy was the same as described above for the Cox model.

All p-values were from 2-sided test, and results were deemed statistically significant at p < 0.05. All analysis were conducted with R version 4.1.3.

Ethical statement

The study was approved by the ethics committees of Madagascar (n° 119 MSANP/CERBM) and the Institutional Review Board of Institut Pasteur (n° IRB/2018/05), France. Written informed consent was obtained for all parents of newborn. If a parent could not read, a witness present during the participant’s information session also countersigned the consent form to confirm the collection of the participant’s informed consent.