Introduction

Globally, an estimated 5.9 million under-five children died in 2015. About 45% of these babies died during the neonatal period (i.e., within first 28 days of life) [1,2]. Though under-five mortality declined by 53% between 1990 and 2015, the neonatal mortality declined slower (47%) than the reduction of post-neonatal under-five mortality (58%) within the same period [1,2]. In 2015, about 98% of neonatal deaths occurred in developing countries, and South Asian countries experienced 40% of these deaths [2]. Most countries in this region have a high neonatal mortality rate (NMR), such as Bangladesh, whose NMR reached about 28 per 1,000 live births in 2014 [2,3].

The Millennium Development Goal 4 (MDG-4) aimed to reduce under-five mortality by two-thirds between 1990 and 2015; many countries failed to achieve this target due to the slower NMR reduction [1,2]. Recognizing the importance of stagnating NMR, the Sustainable Development Goals (SDGs) explicitly stated the need to reduce it. The SDG 3.2 aims to reduce the NMR to less than 12 per 1,000 live births in all countries by 2030 [4]. Although Bangladesh achieved the MDG-4 target by reducing the under-five mortality rate to 46 per 1,000 live births, the progress of NMR reduction has been slower. For instance, from 1990 to 2015, the post-neonatal under-five mortality rate declined by 71% while the NMR decreased by only 46% [3]. In 2014, more than 60% of under-five children deaths in Bangladesh occurred among neonates [3]. Therefore, accelerating the pace of progress for reducing neonatal deaths is vital to achieve the NMR target of SDGs.

Previous studies have concluded that multiple factors simultaneously affect NMR in developing countries including Bangladesh [5–10]. Infants’ individual characteristics such as lower gestational age and gender are associated with greater neonatal deaths [11–13]. Maternal complications during, before, or after delivery, as well as socioeconomic status of the family are also important determinants of neonatal mortality. Prevalence of these risk factors is disproportionately higher in developing countries including Bangladesh; the observed higher NMR results from this disproportionate prevalence in resource-limited settings [5–10].

Studies have also shown that an increased coverage of currently available, feasible, and cost-effective interventions like cord cleansing with chlorhexidine, kangaroo mother care (KMC) or antenatal corticosteroids could reduce about three-fourths of neonatal deaths [14,15]. Identifying risk factors and addressing them through a package of evidence-based interventions are essential to avoiding these unwanted and preventable deaths. Understanding the factors associated with neonatal mortality in rural areas is of particular importance to reduce the high NMR in Bangladesh since over 65% of the population lives in these regions [16]. Moreover, in 2014, the estimated NMR was higher in rural areas (31 per 1,000 live births) in this country compared to urban areas (21 per 1,000 live births) and to the national average (28 per 1,000 live births). Besides this rural-urban difference, a wide regional or divisional variation has also been observed in NMR estimates. For instance, in 2014, the estimated NMR in Sylhet division was 39 per 1,000 live births, while it was 21 per 1,000 live births in Barisal [3]. Although many studies have investigated the rates and risk factors for neonatal or childhood mortality in Bangladesh, there has been limited research about such determinants in rural Sylhet. This lack in epidemiologic literature limits our understanding of the factors to prioritize for an evidence-based programming in this high-risk region of neonatal survival [9,17–22].

Using data from a community-based cohort study conducted in rural Bangladesh, we examined the rates and determinants of neonatal mortality. We assessed the association of individual, maternal, intrapartum, household, and socioeconomic factors with neonatal deaths. Our study may find relevant applications to design interventions aimed at reducing neonatal mortality in low-resource settings.

Methods

Conceptual framework and study variables

From the available theoretical frameworks to address childhood and neonatal mortality, we adopted a conceptual framework from a similar study conducted in Nepal by Paudel et al. [28]. This conceptual framework was adapted based on the data structure of the ANISA study. According to that framework, neonatal mortality is affected by the simultaneous presence of multiple factors. Potential risk factors for neonatal mortality were grouped into the following categories: (i) Infants’ individual characteristics: sex (male and female), gestational age (<34, 34–36, and ≥37 weeks), multiple gestations (yes and no), and birth rank (1^st, 2^nd to 4^th, and 5^th or higher); (ii) Maternal characteristics: birth interval (≤2 and 2–4 years), maternal age (<18, 18–24, 25–29, 30–34, and ≥35 years), maternal mid-upper arm circumference (MUAC) (<22 and ≥22 cm), history of previous child death (yes and no); (iii) Antenatal factors: antenatal complications (pregnancy complications such as high grade fever, excessive bleeding, convulsion, swelling of face or feet, foul smelling discharge or others), number of antenatal care seeking visits (no antenatal visit, 1–3, and ≥4 visits), receiving a tetanus toxoid injection (yes and no), and tobacco consumption during pregnancy (yes and no); (iv) Intrapartum factors: delivery complications such as excessive bleeding, convulsions, retained placenta, abnormal presentation, prolonged labor, premature water breaking, or others (yes and no), place of delivery (hospital and home delivery), delivery assistance (by skilled birth attendants, [doctors and nurses were included as skilled birth attendants]; yes and no), and season of delivery (winter or not winter); and finally (v) Socio-economic and household factors: maternal education (no formal education, primary, and secondary or above), women work for cash (yes and no), women’s decision-making power about treatment of children (yes and no), paternal education (no formal education, primary, and secondary or above), and household economic status (poorest, poorer, middle, richer, and richest) (Fig 1).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Conceptual framework for determinants of neonatal mortality [adopted from Paudel et al[28]].

https://doi.org/10.1371/journal.pone.0206795.g001

Study participants

Fig 2 shows the study profile. Among 28,960 identified pregnant women, 4,689 (16.2%) women were excluded due to refusal (100), false pregnancy (206), abortion (323), administrative censoring before the occurrence of pregnancy termination (2,311), and loss to follow-up (1,749). Of the 24,271 recorded deliveries, 24,560 babies were born including 23,989 singletons, 275 twins, and 7 triplets. After excluding 838 stillbirths, 2 refusals, and 9 babies who were lost to follow-up, the total number of live births was 23,711. As 2,484 newborns were not eligible for follow-up due to the late identification, 21,227 newborns were included in the final analysis.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. Study profile.

https://doi.org/10.1371/journal.pone.0206795.g002

Results

Table 1 shows the NMR according to newborns’ background characteristics. Of the total 21,227 included newborns, 922 babies died within the neonatal period. The overall NMR was 43.4 (95% CI: 39.3–48.0) per 1,000 live births.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Distribution of individual, maternal and socioeconomic characteristics along with NMR (95% CI).

https://doi.org/10.1371/journal.pone.0206795.t001

Results of logistic regression analysis

Table 2 shows the results of simple and multivariable logistic regression analyses along with the model fit statistics. The first model (adjusted for individual factors) had all significant variables. Direction and magnitude of individual factors from the unadjusted level did not change significantly in this model. In Model 2, after adding the maternal & intrapartum characteristics, individual variables did not change significantly from Model 1. The association of maternal age with neonatal mortality attenuated, but one age group remained significantly associated: 30–34 years (AOR: 1.5, 95% CI: 1.2–1.9). Delivery complications (AOR: 2.0, 95% CI: 1.6–2.6) also had greater odds of neonatal deaths.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. Multivariable models for potential risk factors for neonatal mortality.

https://doi.org/10.1371/journal.pone.0206795.t002

After adding the household and socioeconomic factors in Model 3, some variables remained significantly associated with the outcome variable. Compared to singleton babies, the odds of neonatal mortality were more than six times higher among babies born to mothers with multiple gestations (AOR: 6.5, 95% CI: 3.3–12.5). Babies born with gestational age <34 weeks also had higher odds compared to term (i.e., ≥37 weeks) babies (AOR: 5.3, 95% CI: 4.2–6.8). Ordered from the most significant associated factors, neonates with the following characteristics had significantly increased odds of dying: first birth (AOR of 2^nd to 4^th child vs first birth order: 3.0, 95% CI: 1.7–5.4), delivery complications (AOR: 2.1, 95% CI: 1.6–2.7), history of previous child death (AOR: 1.6, 95% CI: 1.2–2.2), and male sex (AOR: 1.4; 95% CI: 1.2–1.7). Although parental education level and household wealth status were associated with NMR in bivariate analyses, none of these factors were significantly associated in the final multivariable model.

The log-likelihood of neonatal mortality increased from Model 1 to Model 3. On the other hand, the AIC decreased from Model 1 to Model 3. The average VIF of Model 3 was 1.10. These model fit statistics including the adjusted-R² indicate that Model 3 explains the variability better than other models.

Discussion

In this community-based cohort study, the burden of neonatal mortality was high at the rate of 43.4 per 1,000 live births. The following factors were associated with higher rate and odds neonatal mortality: male sex, multiple gestations, lower gestational age, first-born child, maternal age of ≥30 years, and a history of antenatal and delivery complications. Although some characteristics such as parental education level and household wealth status were significantly associated with NMR in crude analyses, the adjusted association was not significant for these factors.

The NMR observed in this study was relatively higher than the national average (28 per 1,000 live births) as well as the overall rate in Sylhet Division (39 per 1,000 live births) [3]. This study was conducted in rural regions of Sylhet Division; in addition to a high NMR in Sylhet division, the overall NMR in rural regions is usually higher than the overall rate in Bangladesh. The high NMR we observed could be explained by the fact that this study was conducted in a rural region in the division with the highest NMR in this country. Furthermore, rural women have poor access to health care [3,32], and all maternal and child health indicators are poorer in Sylhet Division compared to other divisions in Bangladesh [2,3].

We found that nearly one-fifth of the babies were preterm, similar to findings from a previous study conducted in this site [33]. Consistent with earlier reports, preterm babies had higher rates and odds of neonatal deaths [9,10,28,34]. Preterm babies have increased risks of infections, hypoglycemia, and hypothermia due to physical immaturity [35]. Globally, prematurity is a leading cause of neonatal deaths [36]. To reduce this high NMR in Bangladesh, the prevention and management of prematurity will be crucial. Several behavioral (e.g., smoking cessation) and medical interventions (e.g., progesterone supplementation) could prevent preterm births [37]. KMC is a useful and low-cost intervention for these babies, as most of the children are born at home without skilled attendants [38,39]. Similar to other studies, we have found an association between neonatal mortality and multiple gestations; this is a well-known risk factor for preterm birth [33,40]. Babies delivered by mothers with multiple pregnancies have greater risks of preterm deliveries as well as antenatal and delivery complications. Simultaneous presence of these factors contributes to higher number of deaths [33,40]. These mothers require adequate birth preparedness to prevent neonatal deaths.

Male neonates had higher rates and odds of dying compared to their female counterparts as has been reported previously [33,41,42]. Male newborns are also more susceptible to infections due to immunodeficiency [43] and have an increased likelihood of congenital malformations [44]. These conditions could increase the risk of death among male newborns. Although the sex of a baby is unmodifiable, this factor is significant from a program planning perspective, in that male infants may require greater attention.

Neonates from mothers with delivery complications such as vaginal bleeding, convulsion or prolonged labor were more likely to die than the neonates born to women without these complications. This finding is common among studies that have examined risk factors for neonate mortality, and indicates that appropriate management of delivery complications would be crucial to ensure better survival of neonates and reduce the overall NMR [12,45–48].

We found an association of birth order with neonatal mortality; first-born children were more likely to die than their next siblings [49]. Prior studies have demonstrated a U-shaped relationship between birth order and neonatal mortality [50]. Mothers of premature infants may not be able to handle sickness of the babies due to inexperience or inability to understand the new responsibilities of child care [11]. Though we did not find teen pregnancy as a risk factor after adjusting for other factors, birth order has been linked with teen pregnancy [11,49]. A significant proportion of primi infants are born to mothers who became pregnant during their teenage years. Along with being inexperienced [11], these mothers perhaps give birth to premature babies because they do not reach full physical or reproductive maturity before pregnancy [49]. There is a need to implement interventions to delay first pregnancy and antenatal interventions to educate first-time mothers.

Our results showed that neonates were more likely to die with a maternal age above 30 years. Consistent with our study findings, a previous study from Bangladesh also found an association between maternal age and neonatal deaths [10]. Parity and maternal age are correlated, as an older mother would have a higher parity than a younger mother, and a mother with a high parity would have higher age compared to a mother with low parity. Earlier studies have observed that neonatal mortality is associated with 5^th or higher birth rank [12,34]. As this is a modifiable factor associated with neonatal mortality, these mothers would require appropriate birth preparedness to ensure neonatal survival.

Women who had previously endured the death of a child were more likely to also experience a death of their newborns in the neonatal period, a finding similarly observed in a case-control study in Bangladesh [51]. Adequate birth preparedness could ensure better neonatal survival among this high-risk group.

Most newborns were delivered at home or by unskilled attendants. Newborns born in health facilities or by skilled attendants had higher NMR compared to newborns born at home or by unskilled attendants. Although this was not a significant factor in multivariable analysis, the higher NMR among these groups could result from complications during childbirth when women were more likely to be delivered in health facilities or by skilled attendants. The mothers utilized these services as ‘curative’ (i.e., to manage delivery complications) services instead of regular or preventive services [26,45].

This study found no association between neonatal mortality and socioeconomic factors; although household wealth status showed a significant association in unadjusted analysis, it was not significant after adjustment. This finding is contrary to those of similar studies from Bangladesh [9,10,20] or other countries [11,12]. However, a recent analysis by McKinnon et al. concluded that the inequalities associated with neonatal mortality are declining in developing countries including Bangladesh, and thus our results are consistent with that finding [52]. Implementation of maternal and neonatal survival programs helped to overcome the socioeconomic barriers of neonatal deaths in these countries [52–54]. The insignificance of the association between socioeconomic and many other factors and NMR indicates that these variables mainly influence neonatal deaths through other intervening factors that were significant such as maternal age, birth order, previous child death, multiple gestations and preterm births. To reduce neonatal deaths, addressing these maternal and intrapartum factors would be important.

Our study has several notable strengths. We analyzed prospectively collected data; recall bias was minimal for pregnancy and delivery complications. Recall bias is a problem of studies which analyze cross-sectional demographic and health survey (DHS) data [9,10,20]. Given the cohort design, we were able to collect data of the women who died due to delivery-related complications while studies that analyzed DHS data could not analyze data of women who die due to pregnancy-related complications; maternal death is closely associated with neonatal death. Our results are generalizable to other rural areas in Bangladesh or in other developing countries. The sample size was large to provide precise estimates. The proportion of missing data was small.

One limitation of our study is that we were unable to include all registered pregnant women (16.2%) or delivered babies (13.6%). Some women were absent on repeated visits, while other women out-migrated during the study period or refused to participate. However, the main reason for not including women in the analysis was that these women did not have a pregnancy outcome during the study follow-up period. Though, these women were unlikely to be different from women who completed pregnancy. Only newborns visited by a CHW within seven days after birth were eligible for ANISA surveillance. Since neonatal deaths occur at a higher rate in the first seven days of life, this exclusion might have led to an underestimation of the burden of neonatal mortality. We relied on the last menstrual period’s (LMP) date to calculate gestational age; it could have led to misclassification of preterm births due to possible recall errors of LMP [33]. Due to lack of data, we could not investigate several other known determinants of neonatal mortality including environmental and genetic factors [11,12].

Conclusion

Our study investigated rates and risk factors for neonatal mortality in rural regions of the division with the highest risk of neonatal deaths in Bangladesh. We found several individual (e.g., sex of the baby), maternal (e.g., maternal age), and intrapartum (e.g., delivery complications) factors that have relationships with higher rates and likelihoods of neonatal mortality. Simultaneous impacts of these factors on neonatal mortality indicate that a comprehensive strategy is required to address them. Prioritizing modifiable risk factors through community-based programs is essential to achieve the NMR target of the SDGs; intrapartum factors are important for this prioritization. Programs need to address reducing the number of preterm births. KMC should be implemented at the community level to manage preterm babies. Adequate birth preparedness is required for ‘primi’ mothers, mothers with multiple pregnancy and with a history of child death. Timely referral and management of antenatal and delivery complications should be ensured. Lastly, more community-based studies are required to determine the feasibility of above-mentioned interventions in rural settings.

Acknowledgments

The Bill & Melinda Gates Foundation (Grant No. OPPGH5307) funded the ANISA study. The authors are thankful to the members of the Projahnmo study team in Bangladesh and the study participants in Sylhet district for their contributions. Projahnmo is a research partnership of Johns Hopkins University (JHU) with the Bangladesh Ministry of Health and Family Welfare (MOHFW) and other Bangladeshi institutions including ICDDR,B, Shimantik, Sylhet M A G Osmani Medical College Hospital (SOMCH), Banghabandhu Shaik Mujib Medical University (BSMMU), and Child Health Research Foundation (CHRF). The other members of the Projahnmo Study Group include: Dr. A. E. Md. Muhiuddin Osmani; Dr. Habibur Rahman, Dr. Mohammad Sharif, Dr. A K M Saiedur Rahman, Dr. Gour Moni Sinha, Md. Kutub Uddin, Dr. Lutfunnahar Jesmin, Dr. Md. Abul Kalam Azad, Dr. Md. Abdullah Al Mehedi, Dr. Md. Moazzem Ali Khan Chowdhury, Dr. Md. Mashuqur Rahman of MOHFW; Prof. Manajjir Ali of SOMC; Prof. Mohammod Shahidullah of BSMMU; Prof. Mathuram Santosham, Dr. Malathi Ram, Dr. Salahuddin Ahmed, Dr. Arunangshu Dutta Roy, Sayed Mamun Moin of JHU; Prof. Nawshad Ahmed, Maksuda Islam, Iftekhar Rafiqullah, Shahidul Islam, Md. Hasanuzzaman of CHRF; and Dr. Ahmed Al Kabir of Shimantik. Abdullah H. Baqui (abaqui@jhu.edu) is the lead author of this group.