A scoping review of methods for assessment of sex differentials in early childhood mortality

Mortality is a result of the interaction between several complex factors such as genetics, environment, health care, and behaviors, and there is overwhelming evidence that male children have a higher risk of death than do females [1‐3]. Nevertheless, there are many historical examples of excess female mortality at specific periods of the life course from childhood through adulthood [4‐7].

Due to biological characteristics, female newborns and infants have an advantage in survival and enjoy lesser vulnerability to mortality due to perinatal conditions, congenital anomalies, and infectious diseases [8‐10]. Beyond the first year of life, girls do not present the same advantage concerning certain infectious diseases, which are the primary causes of death from 1 to 4 years of age, and this advantage remains but tends to decline [1, 2]. Even so, in circumstances in which boys and girls receive the same care, higher mortality among male children is observed and equality in male and female death rates (or higher female mortality) may be evidence of inequity [11].

Excess female mortality may be an indicator of influences that outweigh the biological survival advantages of girls [1]. For instance, in some populations, the female advantage can be eroded if girls are deprived of access to health care or proper nutrition, suggesting discriminatory behaviors and gender bias in child care due to community male preference [1, 12].

Although sex differentials in early childhood mortality are recognized as important issues globally, there is no consensus about what would be expected in the absence of discrimination. A further challenge is that the magnitude of differences will vary according to total mortality levels and to the distribution of causes of death [8, 11, 13].

Many studies have assessed differences in mortality between boys and girls and sets of reference values for the relationship between the sex-specific estimates have been proposed, however, there is not a standard method for such analyses. In particular, the literature still lacks a proper description of how the estimates for such variations had been defined. By bringing together the approaches used so far and using a scoping review methodology, we described how studies addressed sex differentials in early childhood mortality with a special focus on the methods used to measure such discrepancies.

In this review, we described 154 studies published since 1929 that employed quantitative methods to address sex differentials in early childhood mortality. The main challenge in these analyses is how to separate biologically determined sex differences from gender inequities that may affect mortality rates.

The studies included in our review reflect the evolution of knowledge over time across world regions. The early literature (up to the 1980s) is predominantly from high-income countries and aimed at the simple description of sex differentials. These studies established higher mortality among boys and explored potential biological mechanisms for such differences. Parameters derived from such early studies were then used in more recent years to identify populations where girls were more likely to die than would be expected. Both types of studies are relevant because early research was essential for interpreting recent findings.

Biological reasons for the lower mortality among girls include, for instance, the presence of two copies of the X chromosome, whereas male individuals have only one. The X chromosome carries more than a thousand genes and many of them are responsible for immune response, fetal development, and metabolic functions [9]. Additionally, higher odds of congenital abnormalities, lower five-minute Apgar score, need for assistant ventilation, respiratory distress syndrome in the male than in female children are some examples of male frailty [10].

Given the biological disadvantage of male children, usually, gender bias is suspected when the mortality of girls is higher than expected, which may be due to the cultural favoritism of male children resulting in neglect of girls. Although such culturally-driven ‘son preference’ is likely the main reason for higher-than-expected mortality of girls, infanticide may also affect sex ratios, although it seems to be restricted to a few societies [43, 52]; the frequency of infanticide, however, is extremely hard to measure. Whatever the reason, bias against girls is likely to affect the sex ratios, leading to either similar mortality rates for both sexes or higher mortality for girls than for boys [43, 159, 160].

Specifically, the complexities in the analyses include the specification of expected values for sex ratios for a given mortality level, the definition of what constitutes excess mortality of girls or boys, and the availability and quality of the data. In light of these conundrums, selected publications were grouped according to the methodological approach.

Fourteen studies (about 9%) aimed at presenting reference parameters, as well as analytical methods to define the expected relationship between male and female mortality. These publications are quite heterogeneous and rely either on the comparison of populations with similar mortality levels or on historical data from currently high-income populations. The most influential articles in this group include the standard developed by Hill and Upchurch (1995) based on the historical experience of high-income populations and the reference by Alkema et al. (2014) based on recent data from 195 countries [2, 12]. Besides, no statistical testing was performed in some studies, and any magnitude of differences between observed and expected values was considered. For these publications, the comparison with populations with different demographic characteristics arises as an additional limitation.

We have classified reference studies as either descriptive or prescriptive. A previous analysis employed two of the reference methods included in this review in a set of surveys from LMICs to compare their potential for identifying countries with evidence of gender bias in U5MR [14]. The authors found higher values of excess female mortality when using the expected values from Hill and Upchurch [12], a prescriptive approach, than when using Alkema et al. [2] estimates, a descriptive method [74].

Twenty-one publications (14%) were comparative studies that used either external or internal comparison parameters. The comparisons used in analyses included parameters defined by reference studies mentioned above, for example, the Hill and Upchurch standards. Other studies relied on Coale and Demeny’s life table models (1966 and 1983) [157, 158]. Even though these tables were not originally proposed for comparing mortality sex ratios – and therefore were not included in the group of reference studies summarized above – they have been widely used to define expected values. The limitation that arises in these situations is that model life tables incorporate estimates of typical sex differences in mortality for different levels of life expectancy for European populations, rather than for levels of child mortality [12]. Additionally, the estimates differ both according to the model and according to the level within each model [77]. It should be noted that present-day LMICs – the principal focus of the analyses – may have epidemiological and demographic profiles that differ from the settings upon which the models used as the reference were based. Additionally, the definition of the comparison group is often based on a priori, arbitrary judgment of absence of gender discrimination. Even so, these comparisons are useful for monitoring sex ratios and identifying outliers. Some studies combined values from published sex ratios in different populations to define expected values, which is questionable.

Other characteristics of the publications identified in the review are discussed below. Different age ranges for mortality were used in the studies. It is important to highlight that neonatal mortality is heavily influenced by biological factors, being less sensitive to familial behavior. Post-neonatal mortality is more likely to be independent of preexisting medical complications and to happen at home, and more dependent on nurturing care provided by the family. Beyond the age of one year, the female mortality advantage is not marked as is the case for infants, and the effects of discrimination become more evident [1]. Nonetheless, there are many biomedical interventions, such as antibiotics or vaccines, that may prevent infant deaths, and for which access may be affected by gender bias in care-seeking behaviors [161].

In terms of the populations under study, most analyses came out of India, a country known for gender discrimination against females and where various interventions have been used to prevent what has been described as “daughter elimination” [43]. India, Bangladesh, and China were explored in almost half of all studies. In these three countries, there is vast evidence of female disadvantage in survival, regardless of the method used to assess it. Interestingly, few standalone studies were based on other South Asian or Middle Eastern countries, which multi-country studies often single out as presenting gender bias. Works that used data from high-income countries were mainly published up to 1980, all of them were included in the narrative group of studies. Most of these studies were focused on the magnitude of male disadvantage in survival. In the few studies reporting higher mortality among girls, this was only observed beyond the childhood period, particularly during adolescence or adulthood, being related to maternal deaths or those associated with the occupation.

Regarding data sources, civil or vital registration systems are the gold standard, as these record births and deaths on a continuous real-time basis and cover a whole defined population. However, few LMICs have such systems in place with sufficient coverage and quality. Therefore, household surveys collecting information on reproductive histories of women, including births and deaths of their children, constitute the main source of nationally representative child mortality estimates for LMICs. However, they may present some limitations. Since they are based on retrospective self-reports, there is always a chance that the number and sex of children have been misreported. Also, there may be errors in the dating of events and age heaping, the omission of events, and sampling errors. The same limitations apply to census data, which are also frequently used for mortality estimation using indirect methods. Waldron (1983) compared mortality sex ratios estimated from the United Nations life tables (based on vital registration) and from the World Fertility Surveys for 17 countries and found a weak correlation between these estimates, with girls presenting 12% higher mortality than boys in the WFS data, while this difference was only 5% in the life tables data for an overlapping 10-year period [8]. As mentioned above, a limitation of all data sources is that they do not allow a direct assessment of female infanticide, which would be very important for identifying gender discrimination.

The reasons behind son preference and daughter neglect vary from society to society. In some countries, the need to pay dowries for marrying daughters may impact in family’s economies, whereas in other cultures, son preference is attributed more to patriarchal systems and low female autonomy [162]. Also, religious roles that can only be performed by men may lead to the family’s preference for male children [163]. Cultural characteristics and societal values sometimes combine with limitations of resources, which leads families to choose internally which children will benefit [162, 163].

Yet, higher-than-expected male mortality could imply that boys experience mortality to a degree disproportionate to their biological disadvantage. Some evidence is provided by historical changes in sex ratios in high-income countries [11]. This study revealed that the decline in infectious diseases and the relative increase in perinatal causes led male-to-female ratios in infant mortality to increase, after which the improvement of obstetric and neonatal care led to a subsequent decline in the sex ratio [11].

We also noticed a lack of standardization in data aggregation levels. The use of national data, for example, may be problematic as they do not account for subnational variability in household and family characteristics, including behaviors influenced by son-biased values, for instance. Studies from India document important differences among the Northern and Southern states, with less evidence of gender bias in the latter [18‐20, 153].

Studies assessing multiple countries often used data sources covering different periods or failing to study a representative sample of a given world region. Studies of time trends will be more informative than those reporting on single points in time, as sex ratios may vary as a function of historical elements such as war mortality, fertility transitions, and famines.

A related issue, addressed in some articles, is evidence that reductions in excess female mortality observed might be related to increasing pre-natal sex selection practices, supporting the hypothesis that infanticide and neglect of girls are being replaced by sex-selective abortion of female fetuses, contributing to the ‘missing women’ phenomenon.

Some authors argue that population sex ratios would be a more appropriate measure to assess gender bias given that these combine pre- and post-natal discrimination and account for sex-selective migration [82]. Population sex ratios were not covered in this review since we were interested primarily in methods for mortality analyses and assessment of the impact on liveborn children. Nonetheless, a few studies included in this review combined assessments of gender bias in mortality (the so-called ‘post-natal female deficit’) and prenatal sex selection [19, 29, 42, 43, 78].

The investigation of factors associated with sex ratios falls outside of the scope of the present work, but it is noteworthy that many studies complemented their survival analyses with the assessment of potential determinants including health care behaviors [14, 24, 112], health status [82], socioeconomic characteristics [31, 47, 84, 85, 90, 98, 111, 126], family composition [76, 99, 109, 110, 123, 148], fertility choices and differential stopping behavior [20, 107, 126, 141, 148]. Also, the societal position and value of girls and women, including education, power, autonomy, control over resources, and economic and social status, may affect their survival probabilities [83].

The discussion about sex differences in mortality goes beyond childhood and from a lifetime perspective, we point out the debate concerning differentials in life expectancy for males and females and the data sources and methodologies that have been a topic of discussion among researchers [164, 165].

The authors of the studies under review call for further investigation on the causes of unexpected sex differentials and the introduction of more systematic monitoring of sex differences in child mortality. Qualitative research is also necessary to document the unfair distribution of resources and discriminatory treatment of boys and girls. The mechanisms of gender discrimination are complex and multilayered, ranging from deliberate neglect in health-seeking behaviors to bias in resource allocation, which is difficult to assess solely through quantitative analyses. Moreover, additional research is needed to understand the national and regional differences in the economic and societal roles of girls and women, which seems to affect female survival.

Some caveats in the review process must be recognized. As scoping reviews do not assess study quality, some of the literature reviewed may be methodologically flawed. Also, the screening process and data extraction were not performed in duplicate. Nevertheless, our work fills a gap in the literature by summarizing studies carried out over a long-time span, and by contributing to the understanding of analyses on sex differentials in childhood mortality. The main strength of our review is the use of comprehensive combinations of terms for building the search strategy, which produced a list of about 17,000 studies to be screened. Through this search, we were able to identify the huge diversity of journals in distinct fields – such as Epidemiology, Demography, and Economics – that have published on this topic, documenting that gender and survival issues affect multiple aspects of global well-being. Additionally, the visual representation of the words used in the titles of the works reveals the diversity of terminologies, which has presented an extra challenge for the review screening process.

Rather than comparing results that emerged from analyses performed with multiple methods, our review focused on the discussion of methodological approaches for identifying gender bias and addressed the need for standardization of methodologies.

In summary, much effort has been made to measure, interpret, and explain sex differentials in early childhood mortality. Over time, there have been advances regarding data availability and quality, and in the sophistication of statistical methods on the relationship between male and female mortality, allowing definitions of what would be the so-called ‘normal’ and in the identification of unexpected differences. Yet, many authors still fail to account for the role of mortality levels in contributing to sex ratios. The issue of what constitutes the expected values of sex ratios for a given level of mortality is still open to debate but the use of model life tables and historical time series still have a role to play.

To overcome the limitations, future research on this topic should account for overall mortality levels, address different groups of age at death, and account for causes of deaths whenever possible. We also suggest the use of more than one reference value to allow interpretation of findings in light of different parameters, and the use of formal statistical testing that account for the variability of estimates of sex ratios.

Addressing factors possibly associated with the sex differentials and the mechanisms underlying the biased estimates is also crucial and should cover distal as well as proximal factors. Time trends analyses, performed with a consistent methodology, can help understand both the sex differentials in a historical perspective and their relationship with changes in development level, improvements in health care, and advances in the gendered nature of social norms.

The results presented here underscore the need for further work on how to identify appropriate references for evaluating sex differentials on child mortality, leading to the formulation of policy interventions aimed at closing the unfair gender gap in childhood survival. Finally, we reinforce the call of Sustainable Development Goal 17.18 for disaggregation of data to identify more vulnerable groups within countries.