Does breastfeeding protect against infection? In developing countries, the answer clearly is yes.¹ However, in industrialized countries the findings are mixed. Numerous nonprospective studies of industrialized populations in the literature examine the association between breastfeeding and infectious disease; many report a protective effect⁵ whereas others report little or no effect at all.^13,14 Prospective studies show similarly conflicting findings.¹⁵ A 1986 review of the literature concluded that breastfeeding "has at most a minimal protective effect [against infection] in industrialized countries."²⁶

The conflicting results have been attributed to flawed methodologies.^20,23,26 Specific criticisms include failure to (1) collect data prospectively at frequent intervals; (2) control for known confounding variables such as day care, infant age, and indicators of socioeconomic status; (3) clearly define feeding and outcome variables; (4) apply appropriate statistical strategies to a population in which both feeding and exposure to infectious agents change over time; and (5) account for small samples with the potential for unspecified biases.

The objective of our study was to determine if breastfeeding protects US infants against the development of diarrhea and ear infection. If the answer was affirmative, we also wanted to know whether the breastfeeding has to be exclusive to confer these benefits.

METHODS

Sampling Frame

Exclusion Criteria

Data Collection

Nonresponse

For this analysis, we excluded women who failed to complete the demographic questionnaire (n = 60), which left 1743 mother-infant pairs. In any given month, we also excluded infants who received neither breast milk nor formula.

There was a slight attrition in the number of questionnaires completed at each successive month. We compared the demographic characteristics of mothers completing all the questionnaires between months 2 and 7 with those missing at least one questionnaire. The two groups were similar regarding household size, infant gender, and income. They were different concerning education and employment. Mothers with a higher level of education were more likely than less educated mothers to complete all the questionnaires. Similarly, homemakers and professional women were more likely to complete all the questionnaires.

Population Characteristics

Classification of Predictor and Outcome Variables

Table 1. Sample Size by Age of Child and Feeding Category in the Preceding Month [View Table]

Each month, the mother also reported whether her infant had experienced diarrhea or an ear infection in the preceding 2 weeks. Diarrhea was defined for the mother as three or more watery or semiwatery stools in a 24-hour period. Before answering this question, the mother was asked to describe the infant's usual stool. We believe this gave her a reference by which to distinguish between normally loose stools and diarrhea. Ear infection was not predefined.

Confounding Variables

In the full models, we also adjusted for infant gender; maternal education, occupation, and smoking pattern; household size and income; and day care use (Table 2). Maternal education, maternal occupation, household size, and household income are socioeconomic indicators that have known associations with both breastfeeding and infectious disease. Day care use increases the risk of infection because it increases exposure to infectious agents, and it decreases breastfeeding because the infant is physically separated from its mother. Maternal smoking is associated with increased risk of infectious disease as well as decreased lactational capacity,³² and it may be a marker of maternal health behaviors. Data on day care and maternal smoking were collected only in months 3 and 6. Therefore, for outcomes in month 4, the day care and smoking data were imputed from responses given in month 3; for outcomes in months 5 and 7, the day care and smoking data were imputed from responses given in month 6. 

Table 2. Confounding Variables Used in Multivariate Analysis [View Table]

Analysis

We used logistic regression to model the effects of covariates on the odds of experiencing diarrhea and ear infections. Because several months of data were included for each child, it was necessary to account for the correlation of the data from 1 month to the next in the estimation of the regression models. To do this, we used the Generalized Estimating Equations approach^33,34 and assumed that the correlation structure of the data was best captured by stationary 1-dependence. This correlation structure assumes that there is a constant degree of correlation between any two adjacent months of data for a single child (eg, outcomes in month 3 are correlated with outcomes in month 4 to the same extent that outcomes in month 4 are correlated with outcomes in month 5, and so on). We experimented with other correlation structures and found little change in the parameter estimates or standard errors. We performed the analysis using the Statistical Package for Interactive Data Analysis (SPIDA), version 6 (Statistical Computing Laboratory, New South Whales, Australia).

RESULTS

Table 3. Percentage of Infants 2 to 7 Months of Age Reported as Experiencing Diarrhea or Ear Infection, by Feeding Category in the Preceding Month [View Table]

Association between diarrhea and feeding in preceding 1 month.

A dose-response relationship also emerged when we assessed the effect of feeding on ear infections (Fig 2). Again, as the amount of breast milk an infant received decreased, relative to the referent group, the risk of developing an ear infection steadily increased, up to an 80% increased risk among low mixed (P = .003) or formula only (P < .001) infants. When we adjusted the model, the magnitude of the effect diminished but the associations in both the low mixed (P = .02) and formula only (P < .001) groups remained significant.

Association between ear infection and feeding in preceding 1 month.

DISCUSSION

In looking at the relationship between feeding and diarrhea (after accounting for confounders), we found the odds ratios across feeding groups showed a steadily increasing risk of diarrhea as the relative amount of breast milk decreased. A similar pattern of protection was seen for ear infections. When compared with exclusively breastfed infants, infants who received formula only had an 80% increase in their risk of developing diarrhea and a 70% increase in their risk of developing an ear infection.

We view these findings in light of the limitations of this study. Because mothers responded via a mailed survey, some misclassification of diarrhea or ear infections may have occurred. Furthermore, the normally loose stools of breastfed infants may have been misreported as diarrhea. Under both of these circumstances, the bias created would weaken the strength of our association. Thus, our results may underestimate the true magnitude of the relationship between infant feeding and diarrhea or ear infections. Second, feeding choices reflect inherent differences between mothers. We addressed this limitation by controlling for a wide variety of sociodemographic characteristics, but other characteristics that were not measured could confound our results. Third, mothers who participate in mail panel surveys may differ from the general population. As we noted earlier, our study population was of a higher socioeconomic status than the general population. This difference should have also weakened the strength of our association, because we expected the benefits of breastfeeding to be harder to demonstrate in high-socioeconomic status groups. Finally, the sample size of this study did not allow us to examine the effect of feeding infants a small proportion of breast milk relative to formula (eg, <20%). This group was small because it is physiologically difficult for a mother to maintain small numbers of breastfeedings over prolonged periods.

A notable strength of this study is its methodology. By using longitudinal data, we were able to examine infectious disease outcomes in relation to the feeding pattern that immediately preceded them. Because the data were collected month by month, it was possible to obtain specific data on supplemental foods as well as the number of formula feedings and breastfeedings, thus allowing a very precise definition of feeding patterns unaffected by recall bias. Appropriate statistical methods were used to account for the longitudinal nature of the data.

It is our hope that the three main points of this study will serve as tools for the education of pregnant and lactating women in the US. First, breast milk provides protection against selected diseases. This adage is by no means new, but past evidence for protection against infectious disease in developed countries has been less than adequate.^{13,14,18,20,24} In the US, where only 53% of women initiate breastfeeding,³⁵ it is a message that deserves renewed attention and support. Second, supplementing breastfeedings with small amounts of formula does not eliminate the protection afforded by breast milk. Clearly, this is not meant to discourage mothers from practicing exclusive breastfeeding, but, in a society where the mother has to meet many responsibilities in addition to feeding her infant, exclusive breastfeeding may not always be practical. Finally, breastfeeding is not an all-or-none phenomenon; the more breast milk an infant receives in the first 6 months of life, the better.