Ecological study of effect of breast feeding on infant mortality in Latin America

Data on breast feeding

We obtained breastfeeding rates from recent nationally representative surveys for 16 of 36 countries in Latin America. With the exception of Cuba and Chile, 11 12 country data were extracted from the demographic and health surveys from Macro International (Calverton, MD) or reports of the US Centers for Disease Control and Prevention (references available from the authors). We calculated country level rates of breast feeding by age group as infant population weighted averages of age specific feeding rates. We calculated regional and subregional rates as infant population weighted averages of country rates. Infant population figures and subregional and regional classifications were based on those of the United Nations.13

For infants in the first four months of life (aged 0-3 months) the categories were exclusive breast feeding, partial breast feeding, and no breast feeding, and for infants aged 4-11 months (4 months to under 1 year) the categories were partial breast feeding and no breast feeding. Exclusive breast feeding means no other liquids or solids except vitamins, mineral supplements, or medicines. Partial breast feeding means receiving some breast milk, regardless of how much.14 We chose to analyse exclusive breast feeding among infants aged 0-3 months because relatively few older infants in Latin America are exclusively breast fed.14

Mortality data

We used attributable risk methods to calculate the fraction of deaths from diarrhoeal disease and acute respiratory infections that could be prevented by exclusive breast feeding of infants aged 0-3 months and partial breastfeeding during the remainder of infancy. 15 16 Relative risks of death from these diseases for the feeding categories (table 1) were obtained from a reanalysis of published data2 to match the age groups analysed (A Barros, personal communication). Except as noted below, data on all cause and cause specific child and infant mortality and of age distribution of child and infant mortality were regional estimates or other aggregates. We assumed that these estimates applied at country level.

In 1990, the proportion of deaths in children under 5 years old in Latin America from diarrhoeal disease was 18.4% and from acute respiratory infections was 14.1%.17 We estimated the proportion of deaths in children under 5 that occurred in infancy as the ratio of the country specific probability of dying before age 1 year and dying before age 5 years.18 Half of deaths from diarrhoeal diseases in children under 5 occurred during the first year of life, as did three quarters of deaths from acute respiratory infections.19 To predict cause specific mortality at 4 months of age, we plotted estimates of cause specific child mortality by age19 and fitted them by log-log simple least squares regression (figure). The proportions of infant deaths from diarrhoeal disease and acute respiratory infections occurring among infants aged 0-3 months were estimated as 0.62 and 0.75, respectively.

Log-log plot of cumulative cause specific child mortality by age, showing data points for ages 6 months, 1 year, 1.5 years, and 5 years¹⁹ and predicted values at 4 months

• Download figure
• Open in new tab
• Download powerpoint

To calculate the fractions of mortality preventable by breast feeding in each infant age group as a proportion of all cause infant mortality, we multiplied the cause and age specific attributable fractions by the cause specific proportions of all cause mortality for the age group. To estimate the fraction of infant mortality from the studied diseases occurring during the first year of life that was preventable by breast feeding, we calculated a weighted average of the preventable fractions for each infant age group, where weights were the proportions of mortality from diarrhoeal disease or acute respiratory infections occurring in the age groups. We used the same method to calculate the preventable fraction of mortality from both diseases in the first year of life, using the proportions of mortality from both causes as the weights. Subregional and regional results were calculated with subregional and regional population figures13 and infant population weighted averages of rates.

For example, for Bolivia among infants aged 0-3 months, the rate of exclusive breast feeding was 0.60, partial breast feeding 0.38, and no breast feeding 0.02 (see table 2). These rates and the corresponding relative risks for diarrhoeal disease from table 1 (4.1 and 15.1) give an age specific attributable fraction (proportion of preventable mortality) for diarrhoeal disease of:

This figure was multiplied by the age specific fraction of infant mortality from diarrhoeal disease (0.086, calculations not shown) to calculate deaths from diarrhoeal diseases that are preventable by exclusive breast feeding of infants aged 0-3 months as a proportion of infant deaths from all causes (0.59×0.086=0.050). We used a similar procedure to calculate the mortality from diarrhoeal disease preventable by partial breastfeeding for the age group 4-11 months.

To calculate the attributable fraction of mortality from diarrhoeal disease in all infants, we added the attributable fraction of mortality from diarrhoeal disease in infants aged 0-3 months of age (0.59), multiplied by the fraction of mortality from diarrhoeal disease in this age group (0.62), to the attributable fraction of mortality from diarrhoeal disease in infants aged 4-11 months (0.11), multiplied by the fraction of mortality from diarrhoeal disease in that age group (0.38), giving a value of 0.4. The number of deaths from diarrhoeal disease in Bolivia that could be prevented by exclusive breast feeding of infants aged 0-3 months and partial breast feeding during the remainder of infancy was obtained by summing both age groups' preventable fractions (0.050+0.006=0.056) and multiplying the sum by the total number of infant deaths in Bolivia. Analyses were performed with Microsoft Excel 97.

Sensitivity analysis

We allowed some estimates to vary in order to show how subregional and regional results respond to possible errors in estimation or modelling assumptions. The selected variables were breastfeeding rates, relative risks, cause specific proportions of mortality in children under 5, the proportion of deaths in children under 5 that occurred in infancy, cause specific proportions of mortality in children under 5 occurring in infancy, and cause specific proportions of infant mortality occurring during the first four months of life.

We used the spreadsheet program @RISK (Palisade Corporation, Newfield, NY) for the sensitivity analysis. Distributions were defined so that either the mean or the mode corresponded with our estimates, and standard deviations so that the 5th and 95th percentiles bracketed most reasonable variation. Uncertainty is represented by reporting the 5th and 95th percentiles of the distributions of the analysed results.

Accuracy of estimates

Our estimates are both conservative and liberal in some regards. We assumed that the excess risk of mortality is independent across the two infant age groups. However, if breast feeding improves long term nutritional status and immune functioning, children who are breast fed in early infancy may have a lower risk of death thereafter. On the other hand, we ignored competing causes of mortality. Some of the 52 000 infants a year who did not die from diarrhoeal disease or acute respiratory infections would die from other causes.

We assumed full compliance with the specified breastfeeding patterns. Shifting all infants to the described breastfeeding group produces the theoretical maximum possible effect. Studies in Latin America suggest that promotion of breast feeding can realistically give a 25% shift (moving a quarter of infants in each increased risk category to the corresponding no risk category).22^–24 Such a shift would prevent around 13 000 deaths (a quarter of our estimate).

Our estimates of the proportions of infant deaths from diarrhoeal disease and acute respiratory infections occurring during the first four months of life are consistent with observations of age and cause specific mortality in children 19 25 and results from population based survivorship models. 26 27 Although the figures are likely to be good approximations at the regional and subregional level, possible variation across countries is ignored.

The relative risks of death associated with feeding category may also be subject to regional variation. Various estimates have been published of relatives risk, 2 7 28 29 none of which is ideal for this case. We used reanalysed estimates from the Brazilian study2 because it was population based, conducted in one of the countries in the region, investigated the relevant causes of death, applied a consistent protocol for assigning cause of death, excluded infant deaths subject to known or suspected confounding factors, and coded feeding status before the onset of fatal illness. Nevertheless, there may be variations in local risk patterns that we could not account for. Although differences between urban and rural areas could be a source of bias,30 at the aggregate level they are unlikely to affect estimates substantially.21 Another reason for using the Brazilian estimates was that the study controlled for factors such as prematurity, low birth weight, and congenital malformations, which increase the risk of death as well as reducing the probability of breast feeding.31 Surveys of rates of breast feeding, however, do not control for these factors.

These limitations, which are common to this type of study, highlight the need for a sensitivity analysis. However, we found uncertainty ranges to be remarkably stable under varying assumptions, and even the lower bounds indicate that promotion of breast feeding should be prominent in child survival strategies for Latin America.

Cultural differences in ideas about breast feeding should not seriously affect the analysis. Since the demographic and health surveys collect data using internationally accepted indicators, assignment to feeding category does not rely primarily on respondents' understanding of breast feeding but on detailed questions about food and fluid intake.

Implications

Our data are useful for comparing promotion of breast feeding with other child survival strategies. Although our methods are suitable for global analysis, we selected Latin America because it had the highest availability of data. Improved breastfeeding rates are likely to have less effect in areas where higher proportions of infants are breast fed (such as rural Africa) and more effect in areas where early weaning is common (such as South East Asia). Nevertheless, since exclusive breast feeding remains uncommon in many countries, improving rates of exclusive breast feeding in early infancy would probably substantially reduce infant mortality worldwide.

Our analysis suggests that interventions to promote breast feeding should particularly target younger infants and focus on exclusive breast feeding. Programmes such as the Baby Friendly Hospital Initiative have been important in improving the rates of successful initiation of exclusive breast feeding.32 The challenge is to provide adequate and comprehensive support to mothers so that they can continue exclusive breast feeding while coping with the demands and strains of daily life.33