Background
Women of reproductive age in low and middle income countries are more vulnerable to the both communicable and non-communicable diseases than those in high-income countries. Some of the leading causes of death in women globally, which are mostly among women in low-and-middle-income countries (LMICs), are death associated with childbirth, HIV/AIDS, tuberculosis (TB), injuries, cardiovascular disease, obesity, cervical cancer, violence, and depression and suicide [
1]. Women in LMICs account for 99% of maternal death worldwide, up to 87% of global cervical cancer deaths, and suffer from the majority of cases of HIV/AIDS and TB globally [
2‐
5]. The World Health Organization (WHO) has estimated that, over the period 1990–2015, there were substantial reductions in female mortality from tuberculosis (− 32%), HIV/AIDS (− 34%), and maternal mortality (− 47%) as a result of globally targeted intervention programs in low and middle income countries [
2]. By contrast, there have been increases in mortality from other causes, including cervical cancer (+ 9%), road accidents (+ 36%) and suicide (+ 6%) in these countries [
2].
Death of a mother has major consequences for children, as well as for families as a whole, given the important roles of mothers in taking care of family members and contributing to household income [
6,
7]. The effect of a death of a mother on child health outcomes is more devastating when it happens in lower resource settings, where mothers are usually primary caregivers and social support is less widely available [
8]. A number of studies have shown that loss of a mother in early childhood results in a substantial increase in the risk of childhood death, and that the magnitude of the risk of childhood mortality is associated with both age of child at death of a mother and the time since the death of a mother [
9‐
11]. A recent systematic review and meta-analysis reported that the risk of child mortality at < 5 years associated with maternal death was 4·1 (95%CI: 2·4–7.0) [
12]. This recent review provided the first overview on the association between death of a mother and early childhood mortality. However, the mortality risk of children aged < 5 years was estimated by combining the effect estimates/or frequencies reported for more finely stratified age groups. Both the age of the child at death of a mother and age at their own death could be important sources of heterogeneity in the association. Furthermore, the level of impact of death of a mother on childhood mortality could be modulated by many factors, including culture, socioeconomic status, disease burden, health care system, and social support networks, which differ between resource settings and time periods [
13,
14]. A review focusing on lower income settings and considering stratified age groups, has potential to better elucidate the association in these settings.
This systematic review, therefore, aims to address the following research question: in LMICs (also referred to hereafter as lower resource settings), among the general population, for children whose mothers died during their childhood (varying age-intervals of ‘exposure’, up to 18 years of age), what is the risk of childhood mortality relative to those whose mothers were alive throughout the exposure period? In this review, the term “death of a mother” refers to a mother’s death from any cause, for children aged from birth up to a maximum of 18 years.
Methods
The review has been undertaken in accordance with the PRISMA guidelines, and the PRISMA checklist for a report of a sysmatic review and meta-analysis has been used to complete this report (Additional file
4) [
15].
Selection criteria
Studies were included if they: were cohort or nested case-control study in a population of children under 18 years of age living in countries identified as low-income countries, lower-middle-income-countries, or upper-middle-income countries (according to the World Bank’s classification at the time the original studies were conducted) [
16]; compared mortality risk for children according to maternal vital status and reported the effect measure as OR (Odds Ratio), RR (Rate Ratio/Risk Ratio/Relative Risk) or IRR (Incidence Rate Ratio), or HR (Hazard Ratio) using data collected after 1979; were published in English from 1/1/1980 to 31/3/2017. Studies were excluded if they: were editorials or reviews; reported pooled data from individual studies included in this review; or examined only HIV-infected mothers (children of an HIV+ mother are at higher risk of death [
17]) or only pregnancy-related maternal deaths (as restricting to specific causes of mother’s death may introduce an additional source of bias). Papers that reported data collected before 1980 or used data from historical populations were excluded to ensure data are comparable and reliable among different studies. For meta-analyses, we included studies that reported any of the above-mentioned mortality-effect measures and refer to these effect measures as relative risks (Table S1A Inclusion and exclusion criteria and S1B List of excluded studies (Additional file
1)).
Search strategy
We searched for relevant publications published during 1980–2017 on MEDLINE, MEDLINE in-process, Embase, and Global Health databases, which cover the international public health literature, including community health and epidemiology. A complete list of the search terms used for all search strategies is included in Table S2 Search strategy (Additional file
1). Initial searches of databases were completed in August 2016 and subsequently updated by monthly alerts until the end of March 2017. Reference lists of retrieved articles were examined and other websites, including Google, Google Scholar and the WHO website were searched for any additional relevant papers. We examined grey literature, for example annual reports on mortality from the Ministry of Health from countries where original studies were conducted (e.g. Bangladesh, South Africa). However, we found that these types of reports mainly reported the mortality rates of either women or children across different cohorts, whilst we required population-based studies which followed up a cohort of women and their children for a period of time and recorded their vital information during that period to estimate risk of childhood mortality associated with death of a mother. Titles and abstracts of the citations identified by the searches were scanned and clearly irrelevant articles and duplicates were excluded. The full texts of the remaining articles were retrieved for more detailed evaluation and assessment for inclusion against the selection criteria. Literature searching, citation screening and study selection were conducted by the first reviewer (DN) in consultation with the second, very experienced, systematic reviewer (SH). Any issues identified during the selection process by the first reviewer were raised and discussed with the second reviewer. All studies identified for inclusion by the first reviewer were independently assessed by the second reviewer for inclusion.
The two reviewers agreed upon the data to be extracted and a specific data extraction format. For each included study, information was extracted on: (1) study characteristics - study design, setting, population, sample size, exposure, comparator, outcomes follow-up and statistical methods (2) results – risk estimates for exposed and non-exposed and effect measurement (OR, RR, IRR, HR) with 95% confidence interval (CI). If studies reported both the risk of death from all causes and the risk of death from all causes except HIV or tuberculosis, the risk estimates for death due to all causes except HIV or tuberculosis were extracted.
The study characteristics and results of all included studies were extracted independently by both reviewers (DN and SH). Disagreements were resolved by discussion between the two reviewers.
Risk of bias assessment
Sources of bias were assessed independently by two reviewers using a 12-item risk of bias assessment tool developed specifically for cohort studies examining risk factors [
18]. This tool was applied without adaptation and assesses the following potential sources of bias: selection of exposed and non-exposed populations, measurement of exposures, measurement of outcomes, timing of outcome with respect to exposure, adequacy of follow-up, participation rates, completeness of follow-up, accuracy of dates, differences in follow-up, differences in missing exposure data, control of confounding and appropriate inclusion of covariates in analyses.
For the assessment of confounding control, age of the child, the family socioeconomic status, the mother’s education and HIV status in areas of high HIV prevalence, were pre-specified as important confounders. Where there were several different sets of data published for the same cohort, each set of data was assessed.
Summary measure, statistical analysis and data synthesis
Initial review of selected studies showed that the risks of childhood mortality varied according to exposures (age of child at death of a mother and time since death of a mother) and outcome (age at risk of child death). Thus, pooled analyses were stratified by both exposures and outcome. The study exposures were ‘death of a mother’ by child age when mother died (15 papers from 12 studies reported various child age ranges at death of a mother: “0–42 days”, “0–12 months”, “0–5 years”, and “0–10 years”) and ‘death of a mother’ by time since mother died (2 studies reported two time periods since mother died: “<6 months ago” and “≥6 months ago”). Various outcome age-intervals were reported, including: early neonatal (0–7 days), late neonatal (8–28 days), neonatal (0–28 days), post-neonatal (1–6 months; 6–12 months), infant (0–1 year), under-five (0–5 years) and other various age ranges (1–2 years; 2–3 years; 3–4 years; 1–5 years, 5–10 years, 0–10 years, and 0–18 years).
Given all included studies were cohort studies conducted in LMICs from 1980s onwards, random-effects meta-analysis techniques were used to pool effect estimates of the within-strata results reported by individual papers, using STATA version 13.1 [
19]. In forest plot representations, summarystatistics below “mother died” and “mother survived” columns were included for descriptive purposes and were not used to calculate the adjusted effect estimates (which were extracted directly from the study reports). Overall effect estimates of different age strata were not pooled in meta-analyses that included more than one estimate from the same cohort. Additionally, as age of child is a potential confounder, we decided to simply include the extracted effect estimates as reported without combining among different age groups. Because overall death is quite a rare endpoint in children age 0–18, ORs, RRs, IRRs and HRs should be reasonable approximations of each other and are referred to hereafter simply as relative risks (RRs) [
20‐
22].
Heterogeneity was assessed using the Chi
2 test statistic and the I
2 statistic [
15]. Heterogeneity was considered significant if
p < 0.05 for the Chi
2 test statistic and where significant and data permitting for sensitivity analysis, potential sources of heterogeneity were explored. Potential sources of heterogeneity considered were HIV prevalence and risk estimate metrics. We were unable to assess publication bias or small-study effects in this review because of the limited number of studies within homologous outcome/exposure strata (and we did not assess a funnel plot of all effect estimates together because heterogeneity due to variation in the underlying effects being measured is known to induce spurious funnel plot asymmetry) [
23].
Discussion
Our review of 15 articles reporting 13 cohorts of 12 original studies included in qualitative analysis, of which 13 articles reporting 11 cohorts of 11 original studies included in meta-analysis, showed that the relative risks of childhood mortality varied by the age of the child at the time of the mother’s death, time since the mother died, and age-at-risk of child death. Overall, we found strong evidence of an association between death of a mother and childhood mortality in lower resource settings. Compared to overall mortality among children whose mother survived, the risk of mortality associated with a mother’s death appeared to be highest for children in their first 6 months of life (~ 35 times) and this risk decreased (to ~ 28 times) for children up to 1 year of age at the time of a mother’s death. Results of sub-group analyses at different discrete sequential intervals of child age suggested that there was an increased risk of childhood mortality for children up to 18 years among children whose mother died when the child was older than 42 days after birth. We also found that the risk of death was higher in the first 6 months following a mother’s death (~ 4.7 times) than it was after 6 months (~ 2.1 times).
The high risk of mortality among children whose mother died in their first 6 months of age is understandable. Reduced breastfeeding which can cause malnutrition and increased risk of death in neonates and obstetric complications at delivery, such as haemorrhage (which is a leading cause of pregnancy-related maternal mortality and subsequent neonate mortality due to asphyxia) are likely to be major contributors to the higher risk of mortality [
40‐
42]. The risk remained high among children aged 1–2 years whose mother died at any age of the child. It is possible that lack of breast feeding, diarrhoea, respiratory infections, and nutrition disorders might increase the risk of childhood mortality at these ages [
9,
31]. As Ronsmans et al., 2010 and Moucheraud et al., 2015 discussed, in settings where up to 90% of infants were breast fed, the risk of child death following a death of a mother was substantial [
9,
31]. The risk also appeared elevated for children aged 5–10 years and up to 18 years. It is possible that lack of mother’s care in daily life, particularly when children are ill might be associated with risk of childhood mortality at older ages [
9]. Child death up to 12 or 18 years of age, following a mother’s death, might be associated with other factors, including disease, malnutrition, poor hygiene, injuries, and psychological/mental problems [
43,
44]. However, mortality risks for adolescents were reported by three studies only and effects among children who were followed up until aged 18 years could be predominantly driven by child mortality at a younger age. A previous review found that the mortality risk of children under 5 years who lost their mother was more than 4 times higher than the risk of those whose mother survived [
12]; our findings are generally consistent with the previous results. However, our review provides more insight into the association between mother and child mortality in LMICs through a detailed analysis which stratified the results by both outcomes (childhood mortality) and exposures (mother’s death).
Our findings emphasise the critical importance of women in familial outcomes and the importance of health care for women through their child rearing years in lower resource countries. As almost all included studies were conducted in rural and semi-urban areas of low-income countries, lower-middle income countries, and upper-middle-income countries, the findings cannot be directly interpreted in the context of high income settings. Our findings suggested that the relative risk of childhood mortality following a death of a mother was substantially higher for younger children (0–1 year) than for older children (≥ 2–18 years). This can be explained by the numerous factors that affect the stages of child development. For children less than 1 year of age, losing a mother will potentially expose the child to malnutrition and infectious diseases due to lack of breastfeeding and maternal care [
45,
46]. Ronsmans found that among motherless children who died in their first year, 50% died of infectious diseases (diarrhoea and respiratory infections) and 16% died from malnutrition [
9]. As Moucheraud et al., 2015 concluded, in a setting where breastfeeding is a sole source of nutrition for newborns and infants and substitute food sources are unaffordable, inaccessible or unsafe, the risk of child death following a death of a mother will be elevated [
31]. For older children, lack of maternal care during daily life and when ill is likely to drive the increased risk of death, especially in settings where there are no extended family or social support structures for motherless children [
8]. It has been shown that in settings where motherless children can receive care from grandmothers or adopted families, the survival rate of those children was significantly higher than motherless children without such care givers [
47,
48].
This systematic review and meta-analysis has several strengths. We reviewed and synthesized in detail the evidence published on the impact of death of a mother on childhood mortality in lower-resource settings. The review was independently conducted by two reviewers, and we followed best practice in systematic review methodology. The majority (11/12) of included studies were cohort studies, one of the best designs for risk factor studies [
49]. Overall, settings were relatively homogenous among studies, with most conducted in rural or semi-urban areas of lower resource settings. Additionally, to account for the impact of HIV on child mortality, we only included studies that were conducted in the general population, and sensitivity analyses were used to check for the impact of HIV prevalence of the study settings on the pooled estimates. Furthermore, given each of the included papers measured death of a mother and child mortality in various outcome and exposure categories, pooled analyses stratified by these categories were conducted to account for heterogeneity. However, given a number of studies reported several mortality risks at different child ages (or a number of original studies were associated with multiple reports), we did not pool the relative risk if more than one estimate was reported from an original study. Additionally, we conducted careful risk of bias assessment for all included reports. Our assessments found that 57% (9/16) data sets of 15 included reports in the risk of bias assessment had low or medium risk of bias for 12 assessed items. Potential confounders for both death of a mother and childhood mortality, including age of child, family socioeconomic status, and population after and before HIV status, were carefully evaluated when assessing for the risk of bias of each individual report. Regarding the qualities of the extracted data, given all included studies used various methods of survival analysis models to produce the relative risks of childhood mortality according to a death of a mothers, 12 out of 15 reports generated effects sizes from multivariate regression models, whereas 2 reports [
32,
33] generated effects sizes using univariate models. Finally, based on our understanding that the association between death of a mother and childhood mortality would vary by level of country development, our current review only focused on investigating the risk of death of a mother on childhood mortality in lower resource countries.
Our review has some limitations. First, at the study level, given our search was limited in English language, we potentially missed a number of studies in other languages. As a result the number of studies that met the eligible criteria comprised 12 original studies (reported in 15 papers). Second, although study designs, study settings, and time periods over which data were collected were similar among the included studies, there was unexplained variation among studies for some strata. Possible sources of heterogeneity include the differences among country-specific study settings in socioeconomic conditions, culture, disease patterns, and health care systems. Regarding the methodology of the original studies, there was variation in sample sizes, quality and design of the study (different length of follow-up, level of detail reported), and methods used to measure the mortality rates (e.g. some analyses reported rates per 1000 live-births, others reported mortality rate per 1000 child-year or per 100,000 child-days). Given these diversities exist, heterogeneity in pooled analyses is understandable. We presented pooled estimates from meta-analysis even in the presence of significant heterogeneity because it has been argued that such analyses can still be appropriate and useful provided effect estimates from individual studies are in the same direction and random effect methods are used [
50,
51]. Third, based on the available data extracted from included studies, our meta-analyses were conducted using variously reported effect estimates (mortality rate ratio, risk ratio, odd ratio and hazard ratio). However, sensitivity analysis stratifying OR vs. Rate Ratio/Risk Ratio/Hazard Ratio showed insignificant differences. Additionally, for many of the age intervals, pooled estimates were based on only one or two studies. Consequently, readers should be suitably cautious about interpretation and generalisation of such estimates, carefully taking into account the levels of uncertainty implied by confidence intervals. We anticipate that HIV is likely to be confounder of the association between maternal and child mortality. Although at study selection and data extraction stages, studies assessing only mothers with HIV were excluded, we were unable to account for individual mothers with HIV in the included studies. For any pooled estimates with high heterogeneity, sensitivity analysis stratifying for HIV status suggested that there was probably not any association with country HIV status. Another limitation of this review was that most of the 12 included studies (10/ 12) reported the risks of childhood mortality among children whose mother died at any time during their childhood (0–1 year, 0–5 years, 0–18 years), probably due to limited numbers. This approach allows us to see the general impact of the death of a mother on childhood mortality. Although the magnitude of the impact could be seen up to many years later in childhood, it could be argued that most of the mother and child deaths might have been driven by deaths occurring at child birth or postpartum. There were 2 studies reported in 3 articles which analysed risks of child death by multiple discrete sequential ranges of age of a child at the time of the mother’s death. This allows for more precise measuring of the risks of childhood mortality associated with death of a mother occurring at different child ages. Results of sub-analysis based on 2 studies stratified the risks of childhood mortality by multiple discrete sequential ranges of age of a child at the time of the mother’s death were broadly consistent with the main findings and suggested that there was an effect of a mother’s death on child mortality even when the child was older than 42 days after birth. Finally, the impact of death of a mother on childhood mortality is only one of several consequences that we selected to address in the scope of this review. Psychological development, school achievement and risk behaviours of motherless children are outcomes of importance for future research [
52‐
55].
Findings from this review emphasized the importance of the Millennium Development Goals (1990–2015) and Sustainable Development Goals (2016–2030) to reduce maternal mortality as well as childhood mortality globally, since avoiding a death of a mother has immense effects on child mortality. Skilled birth attendants and emergency obstetric care for childbirth complications have been recommended [
56]. For other causes of female deaths, including HIV/AIDS, cancer, cardiovascular diseases, and other chronic diseases, targeted-disease prevention and control programs (many for both males and females) have been recommended under several initiatives. In 2011, at the World Economic Forum, WHO introduced “best buy” interventions which target lifestyle risk factors and disease, including tobacco use, harmful alcohol use, unhealthy diet and physical inactivity, and cancer [
57]. Recently, new Disease Control Priorities, 3rd edition (DCP3) were published by the World Bank, and these aim to assist country decision makers in prioritizing various health issues within limited resources and budgets [
58]. Many of these recommendations are related to the leading causes of female deaths in lower resource countries [
58]. Our findings demonstrate that these programs will not only benefit women, but also provide benefits in terms of familial outcomes.
Publisher’s Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.