Background Studies in South Asia have documented increased risk of neonatal mortality among girls, despite evidence of a biological survival advantage. Associations between gender preference and mortality are cited as reasons for excess mortality among girls. This has not, however, been tested in statistical models.
Methods A secondary analysis of data from a population-based randomised controlled trial of newborn infection prevention conducted in rural southern Nepal was used to estimate sex differences in early and late neonatal mortality, with girls as the reference group. The analysis investigated which underlying biological factors (immutable factors specific to the newborn or his/her mother) and environmental factors (mutable external factors) might explain observed sex differences in mortality.
Results Neonatal mortality was comparable by sex (Ref=girls; OR 1.06, 95% CI 0.92 to 1.22). When stratified by neonatal period, boys were at 20% (OR 1.20, 95% CI 1.02% to 1.42%) greater risk of early and girls at 43% (OR 0.70, 95% CI 0.51% to 0.94%) greater risk of late neonatal mortality. Biological factors, primarily respiratory depression and unconsciousness at birth, explained excess early neonatal mortality among boys. Increased late neonatal mortality among girls was explained by a three-way environmental interaction between ethnicity, sex and prior sibling composition (categorised as primiparous newborns, infants born to families with prior living boys or boys and girls, and infants born to families with only prior living girls).
Conclusions Risk of neonatal mortality inverted between the early and late neonatal periods. Excess risk of early neonatal death among boys was consistent with biological expectations. Excess risk for late neonatal death among girls was not explained by overarching gender preference or preferential care-seeking for boys as hypothesised, but was driven by increased risk among Madeshi girls born to families with only prior girls.
- DEVELOPING COUNTR
Statistics from Altmetric.com
Studies investigating sex differences in neonatal mortality (death during the first 28 days of life) have reported that boys are at approximately 20% greater risk of mortality than girls.1–6 The majority of this research has been conducted in high-income countries, where sex differences are primarily biologically driven and the predominant causes of death are non-infectious.1–6 Studies in South Asia have found that girls sometimes experience greater risk of neonatal mortality, especially during the late neonatal period (days 8–28).7–10 This trend has been explained by differential care-seeking behaviours and gender preference.9–15
Studies have documented biological differences between boys and girls that likely contribute to sex differences in neonatal mortality, favouring survival of girls. Boys are, on average, larger and heavier than girls, a likely important contributor to higher rates of delivery complications and caesarean sections.2 ,6 ,16 ,17 They are more often born prematurely, with less mature lungs at the same gestational age, and more frequently suffer from respiratory morbidity.2 ,5 ,6 ,17 Animal models show that testosterone suppresses the immune system in males, while progesterone and oestradiol enhance innate and humoral immune responses in females.18–20 This is supported by evidence that boys are more susceptible to respiratory and gastrointestinal infections, experience related complications with greater frequency, and more often require hospitalisation throughout their lives.21–24 A surge of testosterone, essential to the development of male sex organs, occurs in utero and for the first 3 months of life, which could impact immune responses during the neonatal period.25
While sex differences in neonatal mortality in high-income countries are primarily attributable to biological differences, the literature points to gender preference and differential care-seeking behaviours to explain observed inversions of risk experienced by girls in South Asia. Studies show associations between sex-specific neonatal mortality, birth order and family composition.12 There are also reports of sex selective termination of pregnancy and infanticide.12 ,26 Additionally, research has shown differences in perceptions of illness by caretakers and care-seeking behaviours, favouring boys.11 ,14 ,27 ,28 Beyond the neonatal period, differences appear in nutritional indicators, preventive and care-seeking behaviours and decisions surrounding expenditure of resources favouring boys.29 Although associations between these factors and mortality have been documented, it has not been shown statistically that these factors explain excess mortality among girls.10 ,14 ,15
Unanswered questions remain regarding the impact that biological (immutable factors specific to the newborn or his/her mother) and environmental factors (mutable external factors) have on sex-specific trends in neonatal mortality. The following secondary analyses of data investigate biological and environmental factors that might explain sex differences in neonatal mortality in an area of rural Nepal, where sex selective pregnancy termination was not available.
Description of the data set
Data for this analysis come from the Newborn Washing Study, conducted by the Nepal Nutrition Intervention Project in 413 sectors (geographic areas) in the Sarlahi District in southern Nepal from September 2002 to January 2006.30–32 This was a pair of nested population-based, cluster-randomised, placebo-controlled trials of antiseptic skin cleansing and newborn umbilical cord care.30–32 The objectives of these trials were to determine whether newborn skin cleansing and/or treatment of the umbilical cord with chlorhexidine reduced neonatal infection and mortality.
These studies have been described previously.30–32 In brief, 23 662 newborns were included in this analysis, of which 17 306 were cluster-randomised by sector to receive a full body placebo or 0.25% chlorhexidine wipe shortly after delivery, 6044 were born after a Data Safety and Monitoring Board (DSMB) recommendation to end the comparative phase of the trial and received chlorhexidine full body wipe, and 312 were enrolled but did not participate in the trial. Of the newborns enrolled in the parent trial, 15 123 were randomised to one of the three umbilical cord care groups: (1) dry cord care (2) soap and water or (3) 4% chlorhexidine. Follow-up of vital status through 28 days was completed on approximately 97% of those enrolled in the study. Vital status data were available for all liveborn infants, regardless of trial participation. Vital status and morbidity data were collected on days 1–4, 6, 8, 10, 12, 14, 21 and 28. There were no differences in impact of chlorhexidine interventions by sex.30 Detailed data were collected on pregnancy and delivery complications, prematurity, birth asphyxia, overall health of the newborn, neonatal infections and birth weight. Information was also collected on newborn care practices via interview at the first visit and on day 14.
Systematic collection of referral and care-seeking behavioural data began on 15 May 2003, as recommended by the DSMB. Based on mothers’ responses and a standardised physical examination, newborns were referred for care if they experienced a prespecified set of symptoms. Care sought since the last visit was documented.
Informed consent and exclusion criteria
Mothers were approached to participate in the study mid-pregnancy and informed consent was obtained orally at that time.32 Newborns were excluded from the trial if the mother declined participation, if the infant died before the first visit, or could not be found at home for weighing and other study visits in the first 10 days of life.32 For the current analysis, however, all liveborn infants of consenting mothers were included.
Data analyses were conducted using Stata V.11.33 Sociodemographic variables were compared by sex and ethnicity, and data were further stratified to compare these variables by sex within each ethnic group to investigate comparability between boys and girls. A crude analysis of sex-specific neonatal mortality was conducted. There were distinct differences between the early (first week of life) and late (days 8–28) neonatal periods, so analyses were conducted separately for the two time periods. Exploratory analyses were carried out using cross tabulations with χ2 and t tests. Sociodemographic data, newborn and maternal characteristics, and biological and environmental factors were compared by sex of the newborn. Environmental factors examined included newborn care practices at birth (delivery, umbilical cord care, bathing, warming and feeding practices), newborn care practices within the first 2 weeks of life (feeding practices, hygiene and skin care practices and warming practices), and care-seeking behaviours. There are two ethnic groups (Pahadi and Madeshi) that reside in the study area. The Pahadi people, of Tibeto-Burman and Aryan decent conform to traditional Hindu caste hierarchy and migrated from the hills of Nepal to the low-lying plains 40–50 years ago. The Madeshi people, with cultural roots in the north Indian-Gangetic floodplain, live along the border between Nepal and Bihar. As there are distinct differences between the groups, analyses were stratified by ethnicity.34–37
As birth weight is typically strongly associated with neonatal mortality, a predictive linear regression model was built to impute missing birth weight values using multiple imputation.38 All birth weights measured more than 72 h after birth, and imputed values based on birth assessments completed more than 72 h after birth were coded as missing. There were 432 birth weights imputed (1.8% of infants born to consenting mothers).
Covariates associated with sex of the newborn and mortality in bivariate analyses were carried forward in multivariate analyses and grouped by biological and environmental factors. Four distinct multivariate models were built; a biological early neonatal model, an environmental early neonatal model, a biological late neonatal model and an environmental late neonatal model. These models were intended to determine which covariates could best explain the relationship between sex of the newborn and mortality, rather than determine which covariates predicted mortality. Generalised estimating equation binomial regression models with a logit link, independent correlation structure, and robust variance were used to estimate ORs to account for the cluster-randomised design, with sex as the primary covariate.
Comparability of data between sexes
There were no meaningful differences between boys and girls in socioeconomic or demographic variables. There were differences between ethnic groups in socioeconomic variables, newborn birth characteristics, newborn care practices and health indicators.
Sex differences in biological factors
Boys were born heavier than girls (boys: 2729 g (SD: 441), girls: 2631 g (SD: 428); Δ=98 g, p value <0.0001) and the prevalence of low birth weight (<2500 g) was statistically significantly higher in girls than in boys (boys: 27.3%, girls: 34.8%; p value <0.0001). Although premature birth (<37 weeks) was, overall, comparable by sex, the prevalence was statistically significantly higher in boys among those who were low birth weight (boys: 30.9%, girls: 26.9%; p value <0.001) and those who were small-for-gestational-age (boys: 9.4%, girls: 7.8%, p value=0.004). Delivery of boys more often required assistance (boys: 73.1%, girls: 71.4%, p value ≤ 0.001), and a larger proportion of boys were delivered in a facility (boys: 9.7%, girls: 8.8%; p value=0.02), although differences were small. Higher proportions of boys had head abnormalities (boys: 0.7%, girls: 0.4%; p value=0.02), experienced respiratory depression (boys: 21.4%, girls: 18.2%; p value<0.001), were unconscious (boys: 7.1%, girls: 5.6%; p value<0.001), and were grey in colour at birth (boys: 3.6%, girls: 2.9%; p value=0.006).
Sex differences in environmental factors
Compared with girls, boys were more often recipients of hygienic practices, received supplemental foods in the first 2 weeks of life, and benefited from warming practices (table 1). Although giving supplemental foods to newborns is often perceived as favourable in these communities it is actually a detrimental practice. While modest, these differences in newborn care practices highlighted an overall trend favouring the care of boys. Additionally, while the proportions of boys and girls referred for care were similar, care was statistically significantly more often sought for boys (table 1). Similar patterns were observed in both ethnic groups.
Crude patterns in neonatal mortality
The overall neonatal mortality rate in this study population was 32.1/1000 live-births, with 76.8% of these deaths occurring in the first week. When stratified by sex, girls had a slight, statistically insignificant, survival advantage over boys (table 2). When further stratified by neonatal period, boys had a 20% greater risk of early, and girls had a 43% greater risk of late neonatal mortality (table 2).
Multivariate analyses: early neonatal mortality
Excess early neonatal mortality among boys was explained by biological factors. Although a wide range of environmental factors were examined, as described in the methods section, there were no measured environmental confounders during this time period. The biological multivariate model included: premature/small for gestational age (SGA) (with imputed missing values and measured birth weight within 72 h of birth), respiratory depression, unconsciousness at or shortly after birth, time of year of birth and an interaction term between sex of the newborn and the time of year of birth (table 3). The interaction term was tested in the model because exploratory analyses showed that birth weights differed by season and bivariate analyses indicated that increases in birth weight changed the relationship between sex of the newborn and mortality. Many of these covariates are in the causal pathway. Treatment groups were initially included in the model. They were neither confounders nor effect modifiers in any of the models presented and are therefore not included as covariates in the final models. There was no evidence that ethnicity modified any of the variables’ effects on the relationship between sex and mortality in the first week of life.
Multivariate analyses: late neonatal mortality
Excess late neonatal mortality among girls was explained by a combination of biological and environmental factors. Low birth weight was the only significant covariate, impacting the relationship between sex of the newborn and mortality, in the biological multivariate model (table 4). Taking into account newborns missing from the multivariate model due to missing covariate data, the crude OR for boys versus girls during the late neonatal period was 0.76 (CI 0.55 to 1.05). Controlling for low birth weight decreased excess late neonatal mortality among girls from 32% to 16%.
The late neonatal environmental multivariate model included prior sibling composition, an interaction between sex of the newborn and prior sibling composition, ethnicity and an interaction between sex of the newborn and ethnicity, and a three-way interaction between sex, sibling composition and ethnicity. The odds of mortality and ORs were calculated from the multivariate model for each of the possible combinations of sex, ethnicity and prior sibling composition (table 5). The largest and only statistically significant difference in mortality rates between boys and girls was observed among Madeshi families with only prior girls.
Patterns in neonatal mortality
The magnitude of the difference observed between boys and girls in overall neonatal mortality was lower than expected. Previous studies have demonstrated that boys are at approximately 20% greater risk of neonatal mortality than girls, compared with 6% observed in this study.2–5 Stratification by neonatal period revealed an inversion of risk between the first and latter 3 weeks of life that made the risk of mortality appear to be comparable between boys and girls during the entire neonatal period. Excess risk of early neonatal mortality among boys was consistent with the literature.2–5 However, the excess late neonatal mortality experienced by girls was divergent from what is observed in high-income countries where most deaths are biologically linked.2–5
Early versus late neonatal period
As hypothesised by Waldron, the data showed that biological factors explained excess early neonatal mortality experienced by boys, and environmental factors best explained excess late neonatal mortality experienced by girls.2 The literature exploring biological differences between boys and girls at birth, showing that boys are more often born with less mature lungs than girls at the same gestational age, and are at greater risk of birth asphyxia and delivery complications, supported the findings of this analysis indicating that respiratory depression and unconsciousness at or shortly after birth had the greatest impact on early neonatal mortality.2 ,5 ,6 ,17 The interaction term between time of year of birth and sex of the newborn indicated that boys born between December and March were at higher risk of mortality than girls. This was the time period when birth weights were highest. It is hypothesised that the increase in birth weight was related to more plentiful food for pregnant women during the harvest and imparted a greater survival advantage for girls than boys during the early neonatal period, increasing the difference in mortality between boys and girls during this time period.
In contrast, during the late neonatal period there were no measured biological factors that could completely explain the excess mortality experienced by girls. While adjusting for low birth weight decreased the magnitude of the association between sex of the newborn and late neonatal mortality, it did not completely explain the excess risk of mortality among girls. Again, it is hypothesised that increased birth weight had a greater impact on survival of girls than boys.
As in other studies, newborn care practices and care-seeking behaviours favoured boys, providing evidence of gender preference.10 ,11 ,13–15 ,27 ,28 Adjusting for these factors, however, did not statistically explain excess late neonatal mortality among girls in this population as many articles suggested they might. On further investigation, the environmental multivariate model showed girls were not at uniformly higher risk of late neonatal mortality. Mortality risk depended on environmental factors such as whether the household was Pahadi or Madeshi and prior sex composition of siblings. Madeshis consistently had a greater risk of mortality than Pahadis, and girls born to families with only girls were at high risk. When infants born to families with only prior girls were removed from the data set, the crude association between sex of the newborn and late neonatal mortality changed from OR 0.70 (95% CI 0.51 to 0.94) to OR 0.84 (95% CI 0.58 to 1.19), rendering the association statistically insignificant. The pattern of mortality observed among girls provides statistical evidence that gender preference among a very specific group of girls, rather than overarching gender preference, explained the inversion in mortality risk in the late neonatal period.
We speculate that the highest risk group for late neonatal mortality (girls born to families with only prior girls) is due to the high value placed on boys in this culture, more so among Madeshi than Pahadi families. Sons not only contribute to household income, but contribute a wife and her associated dowry. Another girl in a family of only girls is perceived as a further burden, as she will require a dowry and will contribute only to her spouse's household after marriage.
Strengths and limitations
These data come from a large population-based study in which vital status was prospectively collected during the neonatal period. Vital status was collected on all live-births, and loss to follow-up was extremely low. A large number of biological and environmental variables were collected allowing in-depth analyses of their impact on the relationship between sex of the newborn and mortality during each time period. As this was a secondary analysis of data, there are additional questions that could not be explored that may further define the results. There may be some survival bias as a large number of neonatal deaths occurred during the first 24 h of life, limiting the amount of biological and environmental information available on these newborns. In particular, biases associated with missing birth weights among those who died very early could only partially be addressed through imputation of birth weights. Birth weight is thus most likely over estimated, as those infants who died during the first 24 h were more likely to be at greater risk for being low birth weight. It is possible that complete data would have further attenuated the excess mortality experienced by boys in the early neonatal period. Since all deaths were accounted for in the primary crude analysis, biases associated with missing data were not a factor in estimation of the initial sex-specific mortality differences.
This analysis highlights three important issues. The first is that neonatal analyses must be stratified by early and late neonatal periods to avoid missing important differences between study groups. Despite common knowledge of significant differences between these time periods in mortality rates, causes of mortality and newborn care practices, many studies continue to evaluate the neonatal period as a whole.
Second, this study provides evidence that biology has a greater impact on early neonatal mortality and environmental factors on late neonatal mortality, as previously hypothesised. Third, gender preference as an explanation for excess mortality among girls during the neonatal period is often over simplified. While studies have provided clear evidence of gender preference, most do not indicate whether, or how, the variables representing gender preference explain excess mortality in statistical models.10 ,13–15 This data set provided evidence of gender preference for boys, but did not show that differences in newborn care practices or care-seeking behaviours impacted the overall relationship between sex of the newborn and mortality during the late neonatal period. Instead, these data demonstrated how a small subpopulation of high-risk girls impacted the risk of mortality for the entire group. These findings more clearly define the role that gender preference plays in divergent sex-specific patterns of neonatal mortality, and provide insight into how to more specifically target interventions in this study population. In populations in which girls are unexpectedly found to be at greater risk of mortality, determining the underlying causes is essential to understanding how to best improve survival in the neonatal period and beyond.
What is already known on this subject
Biologically, boys are at approximately 20% greater risk of mortality than girls during the neonatal period; however, in South Asia some studies have shown girls to be at greater risk of mortality during this time period. This inversion of risk has been explained in the literature by the existence of gender preference and differential care-seeking favouring boys in these populations, but has not been tested in statistical models. This study investigates the biological and environmental factors that might statistically explain the observed sex differences in early and late neonatal mortality in rural Southern Nepal.
What this study adds
Consistent with biological expectations, this study found that boys were at higher risk than girls of early neonatal mortality, but that there was an inversion of risk between the early and late neonatal periods putting girls at increased risk of late neonatal mortality. While exploratory analyses showed differences in newborn care practices and care-seeking favouring boys, these differences did not explain the excess mortality among girls during the late neonatal period in multivariate regression models. Instead, we found the most important factors influencing the inversion of risk during the late neonatal period were ethnicity and prior sibling composition. Girls born to families with only prior living girls were driving the increased risk, and this was especially true within the Madeshi ethnic group. Thus, this study showed that excess mortality among girls was not linked to overarching gender preference within the population, but was linked to a specific group of high-risk girls.
Contributors SR and JK made primary contributions to the design and conduct of this analysis, interpretation of the results and writing of this manuscript. SR had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analyses. JK, GLD, LCM and JMT contributed to the study design, conduct, analysis and interpretation of this and the parent trial results. SKK and SCL contributed to the study design, field conduct, quality control of the parent trial and interpretation of results. All authors have reviewed and approved the manuscript.
Funding This study was supported by the National Institutes of Health (HD 44004, HD 38753); National Eye Institute Training Grant provided through Clinical Trials Training Program in Vision Research (EY 07127); the Bill and Melinda Gates Foundation (810–2054); Cooperative Agreements between JHU and the Office of Health and Nutrition, and US Agency for International Development (HRN-A-00-97-00015-00, GHS-A-00-03-000019-00). Procter and Gamble Company provided commodity support.
Competing interests None.
Ethics approval Committee for Human Research at the Johns Hopkins Bloomberg School of Public Health and by the Nepal Health Research Council. It was registered with http://www.clinicaltrials.gov (trial number: NCT00109616).
Provenance and peer review Not commissioned; externally peer reviewed.
If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.