This study was part of the Bacterial Infections and Antimicrobial Drug Resistant Diseases among Young Children in Low-Income Countries (BIRDY project, http://www.birdyprogram.org). The BIRDY project investigates and responds to consequences of bacterial sepsis and antimicrobial drug resistance in children <2 years of age (protocol in Technical Appendix). The study was authorized by the Institut Pasteur in Paris and by the Ethics Committee in Madagascar. Informed consent was obtained for all participants.

Study Areas and Study Population

Figure 1

Figure 1. Locations of Antananarivo and Moramanga in Madagascar.

The study population included all neonates born in 3 districts (Avaradoha, Besarety, and Soavinadriana) of Antananarivo (the capital of Madagascar, with a catchment area population of 14,997 and 4,128 women of childbearing age) and those of the rural city of Moramanga (catchment area population of 17,159 and 3,795 women of childbearing age) (Figure 1). These areas were chosen because their populations, from poor to extremely poor, were representative of the general population.

Recruitment

Before Birth

We exhaustively identified pregnant women within the study areas during their routine third trimester antenatal visit and pre-enrolled those who met the following criteria: routine residence in the study area with no plan to move away during the follow-up period and no opposition to the research being conducted or to the collection of biological samples (Technical Appendix). We actively monitored preincluded women to ensure enrollment of their neonates at birth. At the time of preinclusion or at delivery, a vaginal swab sample was collected from the pregnant women to detect group B Streptococcus (GBS).

At Birth

To ensure the exhaustiveness of live-birth recruitment, newborns were eligible at birth, even if their mothers had not been pre-enrolled. Neonate inclusion criteria were similar to preinclusion criteria of pregnant women: neonates born to parents living in the study area with no plan to move during the follow-up period; those whose legal guardians were informed and had no objection to the study procedures and collection of biological samples; and those for whom written consent was obtained from at least 1 legal guardian.

We collected fecal samples from the mothers perinatally to test for extended-spectrum β-lactamase (ESBL)–producing Enterobacteriaceae. We also collected the mothers’ sociodemographic, medical, and obstetric characteristics; delivery information; and the neonates’ anthropometric measurements and Apgar scores.

The neonates were examined at birth, and risk factors for infection (Technical Appendix) were assessed. The presence of risk factors for infection led immediately to collection of a placental biopsy sample and collection of gastric fluid (before the first feeding), deep auditory canal samples, and anal swab samples from the neonate to document perinatal bacterial colonization. We then referred neonates with suspected infection to a participating hospital for evaluation. When indicated, antimicrobial drugs were empirically administered according to the World Health Organization (WHO) criteria. For these neonates, we obtained blood samples and lumbar puncture samples (if indicated) beforehand (14).

Incidence of culture-confirmed severe neonatal infections in a community-based cohort of neonates in Madagascar was high (17.7 cases/1,000 live births). These infections are usually difficult to document, especially where women frequently deliver their babies at home, because neonates may show few symptoms before the infections progress rapidly. By using active community recruitment and follow-up, we were able to identify severe neonatal bacterial infections, including those with very early onset. Also, performing blood cultures before initiating antimicrobial drug therapy increased the likelihood of identifying a pathogen.

In low-income countries, incidence estimates for severe neonatal infections are few and the available data are heterogeneous (10). On the basis of community recruitment, Darmstadt et al. estimated an incidence rate of confirmed severe neonatal infection of 2.9 (95% CI 1.9–4.2)/1,000 live births almost 10 years ago in Bangladesh; this rate is much lower than the one we found (17). However, the findings of Darmstadt et al. may be underestimated because of delayed care seeking and a shorter active surveillance period.

Our incidence estimate is lower than the 44.8 early-onset infections/1,000 live births found by Turner et al. on the Thailand–Myanmar border; their estimate was based on a clinical definition of infections and was thus possibly overestimated (18). Also, our incidence risk (1.6%, 16/981) is lower than the pooled incidence risk for possible severe bacterial infection (7.6%) estimated by Seale et al. in a metaanalysis of 22 studies (5). However, the designs of the studies contributing to the Seale analysis varied widely. Our community-based study with pre-enrollment of pregnant women before delivery is likely to reflect a higher ascertainment.

The bacterial isolation rate in blood culture is low (≈10%) for infected neonates in high-income and low-income countries (19). Blood cultures require that trained staff collect these samples before any antimicrobial drug use is initiated. These practices are not routine in low-income countries, and some highly suspected infections could not be bacteriologically confirmed, even in the setting of our research protocol, which included continuous training. However elevated, the incidence rate of confirmed severe infections may therefore be underestimated in our study.

As a comparison, in 2008, the United States reported 0.77 early-onset infections/1,000 live births (20). Although most studies in high-resource settings focus on the early neonatal period and are not population based, our results clearly suggest a much higher burden of neonatal infections in Madagascar than in high-income countries.

Most (75%) neonatal infections occurred during the first week after birth, most during the first 3 days. This finding confirms that community-based active surveillance in the very early period of life is crucial for capturing infections in neonates (4). This result also points out the value of reinforcing interventions and research programs targeting the perinatal period.

Gram-negative bacteria were predominant; the most prevalent pathogen isolated was Klebsiella spp. In a review of studies reporting the etiology of serious bacterial infections in community settings, Zaidi et al. found that Klebsiella spp., E. coli, and S. aureus were the most prevalent bacteria isolated during the first week of life (21–25). In our study, S. aureus was not predominant. It is possible that healthcare workers caring for mothers and neonates in our study were more prone to use clean birth kits distributed by the BIRDY program and to follow good hygiene practices, potentially minimizing horizontal transmission of S. aureus to newborns.

The overall burden of GBS infection in the developing world is not clear; incidence ranges from 0.3 to 0.6 infections/1,000 live births (26). Our study identified no GBS infections. One possible explanation for this low incidence is that several early-onset GBS infections may have not been identified because of rapid death (27). However, this hypothesis is unlikely because we performed close and active surveillance directly after birth and no deaths occurred during the very short period between delivery and neonate enrollment. However, we cannot exclude the possibility that GBS might have been responsible for some cases of infection that could not be bacteriologically confirmed for neonates with clinical signs of sepsis, including some who died. In the context of GBS vaccine development, if confirmed, this low incidence may bring into question the potential cost-effectiveness of maternal vaccination in low-income countries.

We found that the proportion of multidrug-resistant infections was significant (50%, 7/14); 28.6% (4/14) of Enterobacteriaceae were ESBL producers, and 1 of the 2 Staphylococcus spp. isolates was resistant to methicillin. One striking result of our study, however, is the relatively low incidence of antimicrobial drug–resistant infections (≈7.7 infections/1,000 live births). We found no carbapenemase-producing Enterobacteriaceae. In most published studies, assessment of antimicrobial drug resistance at the community level is based on the proportion of resistant infections at hospital admission, which may lead to biased conclusions because of variability in care access and case severity. Our results enabled a more complete picture of this issue and suggest that multidrug-resistant infections in the community may be less problematic than previously estimated.

Nevertheless, more than three quarters (11/14) of the pathogens we isolated were resistant to at least 1 antimicrobial drug recommended by WHO, including 4 isolates resistant to both recommended drugs (14). These findings are consistent with those of several studies conducted in hospital or community settings, which also highlight reduced susceptibility to at least 1 antimicrobial drug recommended for empirical treatment (resistance ranging from 43% to 97%) (19,22,28). In contrast, we observed that the most frequent attitude of physicians in Madagascar was to prescribe 3 antimicrobial drugs, including ampicillin, a third-generation cephalosporin, and gentamicin, when invasive bacterial neonatal infection was suspected. The use of large and unnecessarily broad-spectrum therapy may contribute to increased rates of antimicrobial drug resistance. The development of rapid diagnostic tests to identify pathogens and their antimicrobial drug susceptibility may therefore prevent unnecessary use of broad-spectrum antimicrobial drugs.

Our study has some limitations. We cannot exclude the possibility that 4 pathogens, including one ESBL-producing Enterobacteriaceae, which we documented in the community, might have been acquired in the hospital because the infections occurred after 48 hours in neonates born in a healthcare facility. However, no hospitalization during pregnancy was recorded for the mothers of any of these 4 neonates. Also, because the neonates were enrolled in a research study, their standard of care might have been higher, including better hand hygiene, than that for most of the population. This Hawthorne effect might have induced bias in our results, such as an underestimation of Staphylococcus-associated infections (29). We also probably changed the evolution of these severe bacterial infections by improving early diagnosis and providing better care. These actions might have helped avoid deaths of neonates, which would otherwise have occurred, and contributed to our underestimation of case-fatality ratio.

In conclusion, incidence of bacterial infections among neonates in a community-based cohort in Madagascar was high, although incidence of multidrug-resistant bacterial infections was relatively low. Most of these infections occurred during the first week of life. Our findings suggest that public health measures to decrease deaths from severe bacterial infection among neonates should focus on improving prevention, early diagnosis, and management of infections and prioritizing intervention strategies according to successes with vaccines, clean deliveries, and care of neonates. Current knowledge gaps, including those associated with local burden, bacterial etiology, and antimicrobial drug resistance profiles of severe bacterial infections in low-income countries, prevent us from having a clear picture of the situation. Recently, several international bodies called for global action to combat antimicrobial drug resistance, deemed a “global health security threat” (11,30). Although antimicrobial drug resistance is a real threat, more community data are clearly needed in countries with limited resources so they can select and prioritize effective preventive and treatment strategies to tackle bacterial infections in neonates.