Parental education and inequalities in child mortality: a global systematic review and meta-analysis

Abstract

Background: The educational attainment of parents, particularly mothers, has been associated with lower levels of child mortality, yet there is no consensus on the magnitude of this relationship globally. We aimed to estimate the total reductions in under-5 mortality that are associated with increased maternal and paternal education, during distinct age intervals.

Methods: This study is a comprehensive global systematic review and meta-analysis of all existing studies of the effects of parental education on neonatal, infant, and under-5 child mortality, combined with primary analyses of Demographic and Health Survey (DHS) data. The literature search of seven databases (CINAHL, Embase, MEDLINE, PsycINFO, PubMed, Scopus, and Web of Science) was done between Jan 23 and Feb 8, 2019, and updated on Jan 7, 2021, with no language or publication date restrictions. Teams of independent reviewers assessed each record for its inclusion of individual-level data on parental education and child mortality and excluded articles on the basis of study design and availability of relevant statistics. Full-text screening was done in 15 languages. Data extracted from these studies were combined with primary microdata from the DHS for meta-analyses relating maternal or paternal education with mortality at six age intervals: 0-27 days, 1-11 months, 1-4 years, 0-4 years, 0-11 months, and 1 month to 4 years. Novel mixed-effects meta-regression models were implemented to address heterogeneity in referent and exposure measures among the studies and to adjust for study-level covariates (wealth or income, partner's years of schooling, and sex of the child). This study was registered with PROSPERO (CRD42020141731).

Findings: The systematic review returned 5339 unique records, yielding 186 included studies after exclusions. DHS data were compiled from 114 unique surveys, capturing 3 112 474 livebirths. Data extracted from the systematic review were synthesized together with primary DHS data, for meta-analysis on a total of 300 studies from 92 countries. Both increased maternal and paternal education showed a dose-response relationship linked to reduced under-5 mortality, with maternal education emerging as a stronger predictor. We observed a reduction in under-5 mortality of 31·0% (95% CI 29·0-32·6) for children born to mothers with 12 years of education (ie, completed secondary education) and 17·3% (15·0-18·8) for children born to fathers with 12 years of education, compared with those born to a parent with no education. We also showed that a single additional year of schooling was, on average, associated with a reduction in under-5 mortality of 3·04% (2·82-3·23) for maternal education and 1·57% (1·35-1·72) for paternal education. The association between higher parental education and lower child mortality was significant for both parents at all ages studied and was largest after the first month of life. The meta-analysis framework incorporated uncertainty associated with each individual effect size into the model fitting process, in an effort to decrease the risk of bias introduced by study design and quality.

Interpretation: To our knowledge, this study is the first effort to systematically quantify the transgenerational importance of education for child survival at the global level. The results showed that lower maternal and paternal education are both risk factors for child mortality, even after controlling for other markers of family socioeconomic status. This study provides robust evidence for universal quality education as a mechanism to achieve the Sustainable Development Goal target 3.2 of reducing neonatal and child mortality.

Funding: Research Council of Norway, Bill & Melinda Gates Foundation, and Rockefeller Foundation-Boston University Commission on Social Determinants, Data, and Decision Making (3-D Commission).

Figure 1

Study selection DHS=Demographic and Health Survey.

Figure 2

Mapping of included studies by location and data type This map shows the number of unique sources identified and extracted from the systematic review across all age ranges for each geographical unit. Studies that represented subnational units or cities are mapped here to their parent countries. Colour indicates the type of data source used in the meta-analysis by country, with darker colours indicating a greater number of unique data sources. DHS=Demographic and Health Survey.

Figure 3

Summary of RRs of child mortality by parental education Error bars are 95% CIs. RRs of child mortality are shown for three age intervals: neonatal (1–27 days), post-neonatal infancy (1–11 months), and under-5 childhood (1–4 years). Maternal education and paternal education are shown by completed years of schooling (colours darken with increasing years of education). All levels of parental education were compared with 0 years of education as reference level. RR=relative risk.

Figure 4

RR of under-5 mortality by parent's education (maternal and paternal) and child age (A) These RR curves show fitted average effect sizes in normal space across the full range (0–18 years of parental education) of exposures. (B) This figure shows how the underlying, normalised data were synthesised to produce the RR curves; normalised ln(RR) can be interpreted as the instantaneous slope of the RR curve implied by each study; data are superimposed with a synthesised average effect size; all of this is done separately by age group, and the other age groups estimated in the model are presented in appendix 1 (p 9). DHS=Demographic and Health Survey. RR=relative risk.

Figure 5

Dose–response relationship between parental education and child mortality By displaying the data from figure 4B across the entire exposure range, we are able to examine the monotonicity and linearity of the data. Models are adjusted for wealth or income, the partner's level of education, and sex of the child. RR=relative risk.

Figure 6

Funnel plots of effect sizes extracted in the systematic review Funnel plots show how the effect sizes of RRs from individual studies systematically vary according to the SE of their observations. Because each child age interval has a different average effect size, as estimated by our models, we plotted the residuals against the SE of the observations. The residuals are defined as the normalised RR of the study minus the age-specific fit. Many studies outside of the funnel would indicate study-level heterogeneity and indicate more deviation from the average effect size than would be expected from chance alone. RR=relative risk.