Breastfeeding and Mortality Under 2 Years of Age in Sub-Saharan Africa

Demographic and Health Surveys (DHS) are nationally representative household sample surveys that use a multistage stratified sampling design to collect health and nutrition data for women of reproductive age (15–49 years), their children (0–59 months), and their households in LMICs. Detailed information on the DHS Program has been described elsewhere.¹⁴  The main advantages of the DHS Program include high response rates of the study populations, the national representative sampling frame, high-quality interviewer training, and the standardized questionnaire used for data collection across countries and over time allowing international comparability across populations. DHS data collection procedures were approved by the Inner City Fund international institutional review boards and relevant human subject committees in each country. Oral informed consent was provided by each survey respondent (mothers in our study). We used an anonymous public-use data set. In this study, we included DHS Program data from all countries in sub-Saharan Africa surveyed between 2000 and 2016 to increase the statistical power of analyses, given that mortality in children <2 years of age is a rare event. In Supplemental Figs 2 through 4, we present flow charts of data selection for breastfeeding pattern analysis, breastfeeding history analysis, and breastfeeding duration analysis, respectively. We used data of 217 674 children for breastfeeding pattern analysis, 161 732 children for breastfeeding history analysis, and 104 676 children for breastfeeding duration analysis.

The study population included mothers and their most recent live singleton births in the 2 years preceding the survey. Under-2 mortality (per 1000 live births) was defined as deaths within 24 months after birth. To limit reverse causality on association between breastfeeding pattern in the first 3 days after delivery and <2-year deaths, young children dying in the first 3 days of life were excluded. Infants dying within the first 11 months of life were excluded in the analysis on breastfeeding duration. We grouped under-2 mortality by death time as 4 to 28 days, 1 to 11 months, and 12 to 23 months for breastfeeding pattern analysis.

To assess the breastfeeding pattern, mothers who had a child aged <2 years at the time of survey were asked whether and what they offered to drink or eat to her neonates in the first 3 days after delivery. Breastfeeding pattern was categorized into exclusive, predominant, or partial breastfeeding. Exclusive breastfeeding was defined as feeding only breast milk (from the mother or a wet nurse) and nothing else with the exception of vitamin supplements or medicines. Predominant breastfeeding was defined as feeding of breast milk plus plain water or water-based fluids (sweetened or flavored water, teas, fruit juice, coffee, etc). Partial breastfeeding was defined as feeding breast milk plus animal milk, infant formula, or foods. No breastfeeding was defined as not feeding breast milk either from the mother or a wet nurse. These definitions are consistent with the recommendations from the WHO.¹⁵  Breastfeeding history (none versus ever) and duration were reported by the mother, and the duration was categorized into <3, 3 to 5, and ≥6 months.

Covariates for the main analyses included maternal age at birth and education level; paternal age at birth and education level; sex of the child, birth order, and birth mode (cesarean versus vaginal delivery); household wealth quintile (first quintile [poorest] to fifth quintile [richest] within each country) and residence (urban or rural); country; and survey year. Smoking status and occupation of parents, antenatal visits and birth size were not consistently collected in several surveys and these covariates were adjusted only in the sensitivity analyses.

The DHS Program uses a complex survey design for participant selection, and clustering, strata, and sampling weights were provided for each survey across countries. To make each survey unique in the pooled data set, the “survey” variable was considered.¹⁶  The recalculated clustering, strata, and sampling weights in consideration of the survey variable were used for calculation of overall under-2 mortality prevalence and in the multivariable logistic regression analyses. Kaplan-Meier survival curves were performed according to breastfeeding pattern. We used multivariable logistic regression analysis to examine the effects of breastfeeding pattern, history, and duration on under-2 mortality with adjustment for the mentioned covariates. Subgroup analyses were performed by categories of <2-year death age (4–28 days, 1–11 months, and 12–23 months). We calculated attributable cases to estimate the numbers of infants who could have been saved if women offered exclusive breastfeeding or along other breastfeeding scenarios. With the relative risk (RR) being estimated by the adjusted odds ratio (OR), the etiologic fraction was first calculated as (RR−1)/RR. The numbers of attributable cases were then calculated by multiplying the etiologic fraction by the deaths in the specific exposure groups.

Several sensitivity analyses were performed to test the robustness of our findings. First, other covariates, including smoking status and occupation of parents, antenatal visits, and birth size, were further adjusted separately and simultaneously. Second, the gross domestic product per capita on purchasing power parity (PPP) (current international dollar) at country level in the survey year was further adjusted.

All statistical analyses were performed by using SAS 9.4 (SAS Institute, Inc, Cary, NC). A P value <.05 was considered statistically significant.

Data on breastfeeding pattern in the first 3 days of life and under-2 mortality (4 days–23 months) were available from 75 DHS surveys in 35 countries in sub-Saharan Africa between 2000 and 2016. The total sample size was 217 112 children, ranging from 574 (0.3%) in Swaziland to 22 649 (10.4%) in Nigeria (Supplemental Table 5). The weighted prevalence of under-2 mortality (4 days–23 months) was 32.1‰, ranging between 11.9‰ in Angola and 45.7‰ in Swaziland. Overall, the weighted proportions of exclusive, predominant, partial, and no breastfeeding were 59.8%, 31.0%, 8.3%, and 0.9%, respectively (Supplemental Table 6). The characteristics of mothers, fathers, offspring, and households in the pooled data set by breastfeeding pattern are presented in Table 1.

In the pooled analysis, compared with children aged 4 days to 23 months exclusively breastfed in the first 3 days of life, those who were not breastfed, predominantly breastfed, and partially breastfed were more likely to die during the first 2 years of life (Fig 1). In multivariable logistic regression analysis, adjusted for potential covariates, compared with children aged 4 days to 23 months exclusively breastfed, those who were not breastfed had a higher risk of mortality during the first 2 years of life (OR = 13.45; 95% confidence interval [CI] = 11.43–15.83) and in each of the examined periods (OR = 23.97, 95% CI = 18.94–30.34 in the 4–28-day period; OR = 9.52, 95% CI = 7.68–11.79 in the 1–11-month period; OR = 8.19, 95% CI = 5.13–13.09 in the 12–23-month period) suggesting that the association was still high but attenuated over time (Table 2). In addition, children aged 4 days to 23 months predominantly breastfed or partially breastfed were also at increased risk of mortality before the age of 2 years compared with children exclusively breastfed (OR = 1.11, 95% CI = 1.03–1.21 for predominant pattern; OR = 1.12, 95% CI = 0.99–1.27 for partial pattern). However, a statistically significant association was observed only in the age subgroup of 1 to 11 months in the analysis by age subgroup (Table 2).

Because the association of predominant and partial breastfeeding patterns on under-2 mortality was larger in the 1- to 11-month age category as compared to the other 2 periods, we performed sensitivity analyses restricted to this period, which revealed consistent results (Supplemental Table 7).

For the association of breastfeeding history (in the first 6 months of life) with under-2 mortality (6–23 months), the total sample size was 161 322 children, ranging from 440 (0.3%) in Swaziland to 16 668 (10.3%) in Nigeria (Supplemental Table 5). The weighted under-2 mortality prevalence (6–23 months) was 18.0‰, ranging from 3.0‰ in Comoros to 25.7‰ in Swaziland. Overall, the proportion of ever breastfeeding was 99.0%. The characteristics of mothers, fathers, offspring, and households in the pooled data set by breastfeeding history categories are presented in Supplemental Table 8.

In the pooled analysis (Table 3), after adjustment for potential covariates, compared with children aged 6 to 23 months ever breastfed within the first 6 months of life, those who were not breastfed had increased risk of mortality before the age of 2 years (OR = 5.65; 95% CI = 4.27–7.47). Subgroup analysis by death time revealed that the association was consistently significant in the 6- to 11-month period (OR = 5.07; 95% CI = 3.66–7.04) and in the 12- to 23-month period (OR = 6.38; 95% CI = 3.94–10.32). Other sensitivity analyses considering other covariates revealed similar findings (Supplemental Table 9).

For the association of breastfeeding duration with mortality during the second year of life (12–23 months), the total sample size was 104 427 children, ranging from 292 (0.3%) in Sao Tome and Principe to 10 656 (10.2%) in Nigeria (Supplemental Table 5). The weighted prevalence of mortality during the second year of life (12–23 months) was 10.6‰, ranging between 0.0‰ in Sao Tome and Principe and 15.9‰ in Niger. Overall, the proportions of no breastfeeding and breastfeeding duration of <3, 3 to 5, and ≥6 months were 1.1%, 0.7%, 0.6%, and 97.6%, respectively. The characteristics of mothers, fathers, offspring, and households in the pooled data set by breastfeeding duration are presented in Supplemental Table 10.

In the pooled analysis (Table 4), after adjustment for potential covariates, compared with children aged 12 to 23 months breastfed for ≥6 months, those who were breastfed for shorter periods were at increased risk of mortality during the second year of life (OR = 2.78, 95% CI = 1.45–5.32 for duration of <3 months; OR = 5.28, 95% CI = 3.24–8.61 for those who were not breastfed). Sensitivity analyses considering other covariates revealed similar findings (Supplemental Table 11).

Using population-based data (N = 217 112 children) in 35 countries in sub-Saharan Africa, we found that (1) compared to exclusive breastfeeding, suboptimal breastfeeding (predominant or partial pattern) and no breastfeeding significantly increased the risk of mortality of young children aged <2 years; (2) ever breastfeeding decreased under-2 mortality of children aged 6 to 23 months when compared to no breastfeeding; and (3) there was a dose-response association between breastfeeding duration and risk of mortality during the second year of life (age 12–23 months). These findings are consistent with the WHO recommendations for exclusive breastfeeding in the first 6 months of life and continued breastfeeding up to the age of 2 years.

Authors of several studies have assessed the association between breastfeeding pattern and offspring mortality in some countries in Africa, but most studies came from only one center or had limited statistical power to provide robust finding, given that under-2 mortality is a fairly infrequent outcome.^7–11  To our knowledge, there are no related data that are focused entirely on Africa. One previous study including 1677 infants who were followed from birth to 12 months of age from Bangladesh revealed that exclusive breastfeeding in the first few months of life was associated with all-cause mortality as compared to partial or no breastfeeding (OR = 2.23; 95% CI = 1.45–3.44) and, in particular, increased mortality due to respiratory infections (OR = 2.40; 95% CI = 1.14–5.04) and diarrhea (OR = 3.94; 95% CI = 1.47–10.57). This study also revealed that predominant breastfeeding was not associated with under-2 mortality (all P > .05).⁹  Similarly, another study including 9424 infants who were enrolled at the age of 18 to 42 days from Ghana, India, and Peru revealed the increased mortality between the ages of 6 weeks and 6 months associated with partial breastfeeding (hazard ratio [HR] = 2.46; 95% CI = 1.44–4.18) or no breastfeeding (HR = 10.5; 95% CI = 5.0–22.0) versus exclusively breastfeeding (HR = 1.46; 95% CI = 0.75–2.86) when compared to predominantly breastfeeding.⁷  In contrast, 1 subsequent meta-analysis including only 3 studies (N = 22 048 children) revealed that compared to exclusive breastfeeding, no breastfeeding was associated with the highest risk of mortality in offspring aged 0 to 5 months (OR = 14.40; 95% CI = 6.13–33.90), followed by partial breastfeeding pattern (OR = 2.84; 95% CI = 1.63–4.97) and predominant breastfeeding pattern (OR = 1.48; 95% CI = 1.13–1.92).¹⁷  Estimates from this meta-analysis were consistent with results from a recent large prospective study including 99 938 young children.¹¹  The results of our study with a large sample size (N = 217 674 children) are consistent with this previous meta-analysis. However, these previous other studies were only focused on deaths within the first 6 months of life, whereas our study reveals that partial or predominant breastfeeding pattern also increases risk of death during the period of 1 to 11 months.

Several previous studies have examined the association between breastfeeding history (none versus ever) and risk of under-2 mortality,^18–23  but the statistical power was generally insufficient, and control for confounders was limited in most of these studies. One previous meta-analysis revealed that compared with children ever breastfed within 6 months, those not breastfed were at increased risk of death (OR = 1.76, 95% CI = 1.28–2.41 among infants aged 6–11 months; OR = 1.97, 95% CI = 1.45–2.67 among young children aged 12–23 months),¹⁷  which is consistent with our findings among 161 732 African young children. However, the magnitude of our estimated associations was larger (OR = 5.07, 95% CI = 3.66–7.04 for infants aged 6–11 months; OR = 6.38, 95% CI = 3.94–10.32 for young children aged 12–23 months) than those of that meta-analysis.

To our knowledge, few studies have evaluated the effect of breastfeeding duration on infant mortality. Data from one US study of ∼6000 postneonates (28 days–1 year of age) revealed that longer breastfeeding was associated with lower risk of mortality, with an OR (95% CI) of 0.62 (0.46–0.82) for 3 months or more of breastfeeding.¹²  Data from the US National Health and Nutrition Examination Survey III of 2277 children aged 6 to 23 months revealed that prolonged breastfeeding decreased the risk of respiratory tract infections, including pneumonia,¹³  which is an important risk factor for under-2 mortality. Our study also supports the notion that prolonged breastfeeding is associated with reduced under-2 mortality, at least in sub-Saharan Africa.

Our findings emphasize that policy makers, health workers, and mothers in sub-Saharan Africa (and likely in all countries) should promote exclusive breastfeeding during the first 6 months of life and prolonged exclusive or partial breastfeeding up to the age of 2 years. In view of the magnitude of the association between breastfeeding and under-2 mortality, adequate breastfeeding holds the perspective to reduce substantially the currently high under-2 mortality in LMICs. Mechanisms through which breastfeeding is beneficial have been widely described,^24–26  including exclusive breastfeeding necessarily decreasing exposure to unsafe water and fecal contaminants, which may be associated with the risk of infant mortality due to infectious disease.

Our study has several strengths. First, to our knowledge, our study has the largest study sample (N = 217 112) for assessing the associations of breastfeeding pattern, history, and duration on under-2 mortality, allowing sufficient statistical power to draw solid conclusions. Second, the use of national representative and comparable samples made our findings generalizable to populations in sub-Saharan Africa and, likely, in other LMICs. Third, we adjusted our analyses for several potential confounders and performed several sensitivity analyses, which makes our findings robust. However, several limitations should also be noted. First, there might be reverse causality between breastfeeding pattern and duration and under-2 mortality. Infants who have severe illnesses may not be adequately breastfed or weaned in early life. This is why we excluded infants who died in the first 3 days of life when assessing the association of breastfeeding pattern on under-2 mortality. In addition, infants who died within the first 12 months of life were excluded when assessing the effect of breastfeeding duration on mortality at the age of 12 to 23 months. These attempts have likely significantly reduced the possibility of a bias due to reverse causality. Second, information on breastfeeding pattern, history, and duration was reported by the mother, and recall biases might underestimate or overestimate the strength of our observed associations. Third, although several potential confounders were adjusted for in our analyses, residual confounding may not be fully ruled out. Fourth, meta-regression analysis can be better than pooling diverse data. We did perform a regression analysis in the data from each country, but we cannot calculate reliable ORs because the under-2 mortality is a rare event in the population and the statistical power of analyses from most country studies is too low.

Our results reveal that exclusive breastfeeding during the first 6 months, as well as prolonged breastfeeding up to the age of 2 years, was associated with substantially decreased risk of under-2 mortality. Of note, 682 (22%) children’s lives could have been saved if all mothers had breastfed their infants for 6 months. These findings support the promotion of exclusive breastfeeding up to the age of 6 months and prolonged breastfeeding up to the age of 2 years as an important mean to reduce the burden of under-2 mortality in LMICs, where such mortality is still much higher than in other regions.

We thank the DHS Program for sharing the available data.

CI

confidence interval

DHS

Demographic and Health Surveys

HR

hazard ratio

LMIC

low- and middle-income country

OR

odds ratio

PPP

purchasing power parity

RR

relative risk

WHO

World Health Organization