Intended for healthcare professionals


Impact of breast feeding on admission for pneumonia during postneonatal period in Brazil: nested case-control study

BMJ 1999; 318 doi: (Published 15 May 1999) Cite this as: BMJ 1999;318:1316
  1. Juraci A César, assistant professor (juraci.cesar{at},
  2. Cesar G Victora, professorb,
  3. Fernando C Barros, professorb,
  4. Iná S Santos, associate professorb,
  5. José A Flores, senior radiologistc
  1. aDepartamento Materno Infantil, Fundaçäo Universidade do Rio Grande, Rio Grande do Sol, Brazil
  2. bDepartamento de Medicina Social, Universidade Federal de Pelotas, Rio Grande do Sol
  3. cHospital Pediátrico Santo Antônio, Porto Alegre, Rio Grande do Sol
  1. Correspondence to: Professor J A César, Maternal and Child Epidemiology Unit, London School of Hygiene and Tropical Medicine, 49-51 Bedford Square, London WC1B 3DP
  • Accepted 26 January 1999


Objective: To determine whether breast feeding protects infants against pneumonia and whether the protection varies with age.

Design: Nested case-control study.

Setting: Pelotas, southern Brazil.

Subjects: Cases were 152 infants aged 28-364 days who had been admitted to hospital for pneumonia. Controls were 2391cases in a population based case-control study.

Main outcome measure: Odds ratio of admission for pneumonia according to type of milk consumed (breast milk alone, breast and formula milk, or formula milk and other fluids only), use of fluid supplements apart from formula milk, and use of solid supplements.

Results: Infants who were not being breast fed were 17 times more likely than those being breast fed without formula milk to be admitted to hospital for pneumonia (95% confidence interval 7.7 to 36.0). This relative risk was 61 (19.0 to 195.5) for children under 3 months old, decreasing to 10 (2.8 to 36.2) thereafter. Supplementation with solids was associated with a relative risk of 13.4 (7.6 to 23.5) for all infants and 175 (21.8to 1405.1) for those under 3 months old.

Conclusion: Breast feeding protects young children against pneumonia, especially in the first months of life. These results may be used for targeting intervention campaigns at the most vulnerable age groups.

Key messages

  • Pneumonia is the leading cause of death in children under 5 years old across the world

  • In Brazil infants who were not breast fed were 17 times more likely than those receiving breast milk alone to be admitted for pneumonia

  • The relative risk of admission was 61 for children under 3 months of age, decreasing to 10 thereafter

  • Supplementation with solids was associated with a relative risk of 13.4 for all infants

  • Mothers must be encouraged to breast feed very young infants and be advised of the right time to introduce supplementary foods


Pneumonia is the leading cause of death in children under 5 years old worldwide, 1 2 and breast milk is the most important food in the first year of life.3

Several studies in less developed countries have assessed the effect of breast feeding on the risk of developing acute lower respiratory infections, particularly pneumonia.4 Most of these studies show a protective effect of breast milk on pneumonia, but causality has not yet been shown.4 In addition, whether this protection changes with age, as has been shown for diarrhoea,5 is not known.

We performed a nested case-control study in southern Brazil to assess whether breast feeding protects young children against pneumonia and whether this protection varies with age.

Participants and methods

Study population

Throughout 1993 all women who lived in urban areas and had their babies in Pelotas, southern Brazil, were interviewed soon after delivery in the city's hospital. Over 99% of all births in this city take place in such hospitals.6 A systematic sample of 655 newborn infants was selected for home visiting at 1 and 3 months of age according to date and time of birth. In the first month the mothers of 99.1% (5256) of the children were interviewed and in the third month 98.3% (5214). These infants were also visited at the age of 6 months. As additional funds were obtained for data collection, the sampling fraction was increased to 1144,including the 655 visited at 1 and 3 months of age. These samples represent 12.3% (655) of all children from the original cohort in the first and third months and 21.6% (1144) in the sixth.

Defining cases

Cases of pneumonia were identified through daily visits to the city's hospitals. Children who were born in 1993 and had been admitted to a hospital when aged 28-364 days were considered for inclusion in the study. Two independent referees (paediatricians) reviewed all the available information on each child from hospital's records. Pneumonia was diagnosed from the presence of all clinical signs (difficult or rapid breathing, chest indrawing), presence of rales, and—whenever available—results of laboratory and radiological tests. Whenever the two referees disagreed, a third senior referee established the final diagnosis.

Defining controls

The control group was made up of children taking part in the cohort study. For cases aged 28-89 days, controls were infants at the first home visit, who were aged about 30 days. For cases aged 90-179 days, controls were infants at the second home visit, and for cases aged 180-364 days, controls were infants at the third visit.

For each interview all available controls were used. The study was therefore ratified for age but not matched at the individual level. A child who became a case at, say, 9 months old should have been a control at an earlier age. This characterises this study as a case base, or inclusive, design.7


The mothers of cases were interviewed at home soon after the infant had been discharged from hospital using the same questionnaire as was used for the mothers of controls. Information on diet was collected for cases at the age of their corresponding controls at the home visit. For example, for a case aged 45 days dietary information was obtained for the exact age of 30 days. Three variables were studied.


Hierarchical framework for multiple logistic regression

Table 1

Unconditional multiple logistic regression model for risk of postneonatal pneumonia

View this table:
  • Type of milk consumed—breast milk alone, breast and formula milk, or other fluids alone (water, teas, juices, formula milk, or any other liquid supplement except breast milk; this group was considered to be completely weaned)

  • Use of fluid supplements—whether infants received water, teas, juices, or any other liquid supplements excluding formula milk

  • Use of solid and semisolid supplements.

Social class was based on family income, parental schooling, the occupation of the head of household (person with highest salary). This resulted in the following categories: bourgeois and new small bourgeois (professionals and owners of large businesses), small traditional bourgeois (owners of small businesses and shopkeepers), atypical proletariat (non-manual workers in regular employment), typical proletariat (manual workers in regular employment), and subproletariat (unemployed and casual workers).8 Family income was defined as the total amount received by all people who lived in the one home during the previous month. This total was converted into the number of minimum wages.

Sample size

The sample size studied was sufficient for detecting an odds ratio of 2.0 for exposures present in 25% of the control children, with an α error of 0.05 and a power of 80%.9 An additional 40% was added to adjust for confounding variables and to compensate for possible refusals.10 According to this estimate, the final sample size should have at least 143 cases and 572 controls (four controls per case).

Statistical analysis

We measured odds ratios with 95% confidence intervals and used the χ2 test for contingency tables in analyses.11 We adjusted analyses using unconditional multiple logistic regression according to a previously determined hierarchical framework (figure). In this model some variables are assumed to mediate their effects through other variables as well as directly. The outcome variable was admission for pneumonia. The significance level for the inclusion of each variable in the model was measured by the likelihood ratio test. The final model included all variables that had a P value up to 0.10 after adjustment for variables in the same and higher levels of the framework. In addition, each ordinal variable—for example, family income group—was evaluated for linear tendency. When this association was significant and did not deviate from linearity, the variable was included in the model as a linear component. When missing values were less than 5% of all cases, they were recorded as the mode. Data on social class were missing in 7% of interviews,12 and we therefore created a separate category for missing values for this variable. All data were analysed with SPSS for Windows13 and Egret.14


Of the 5304 infants in the original cohort, 152 (2.9%) were admitted to a hospital with pneumonia in the postneonatal period.

Among 250 variables tested, only social class, family income, and maternal schooling, age, parity, and weight gained during pregnancy were associated with outcome (table 1).

Table 2 shows the frequency distributions of cases and controls according to these variables. The fact that there were more controls aged 6-11 months was due to the sampling scheme used in the cohort study. This does not affect the analyses since all information on feeding was referred to the exact age at the start of each age range.7

Table 2

Numbers (percentages) of cases of pneumonia and controls according to main risk factors

View this table:

Table 3 shows that the relative risk of admission for pneumonia for infants receiving breast and formula milk or other fluids alone was 3.8 and 16.7 respectively in comparison with infants who were exclusively breast fed. When infants who received fluid supplements were compared with those who did not the risk disappeared after adjusted analysis. Infants receiving solid and semisolid supplements had a relative risk of 8.5 of being admitted in comparison with those who did not receive such supplements.

Table 3

Odds ratios for developing pneumonia according to type of food given

View this table:

Table 4 shows that after adjustment infants receiving breast and formula milk at the age of 1-2.9 months were 2.9 times more likely to be admitted for pneumonia than were those who received breast milk alone. The relative risk for infants who were completely weaned was 61.1. From age 3-5.9 months these relative risks were 3.4 and to 10.1 respectively. From age 6-11.9 months the odds ratios were 3.7 and 9.2 respectively. The interaction between age and the type of milk consumed was significant(P<0.001).

Table 4

Odds ratios for developing pneumonia according to type of food given stratified for age

View this table:

The crude analysis showed that the risk associated with the intake of supplementary foods in the first months was 175 for children aged 1-2.9 months, 9.1 for children aged 3-5.9 months, and 0.7 for children aged 6-11.9 months. The odds ratio of pneumonia admission for all children who received supplementary food was 13.4.


Methodological limitations

Case-control studies may be affected by several biases. 4 1518 Reverse causality bias—that is, repeated respiratory illnesses leading to a change in breastfeeding pattern—was avoided by regarding as still breastfed infants who had stopped breast feeding because of a respiratory infection up to two months before admission. Another possibility is recall bias, since mothers of cases in a given age range (1-2.9, 3-5.9, and 6-11.9 months) were asked to provide retrospective information on feeding patterns at the beginning of that interval, while mothers of control children were interviewed within a few days of that date. To assess how this could affect the estimates of relative risk we analysed the reported feeding patterns of 32 infants who had been both a case and a control. For 26 infants the type of milk consumed was the same in both interviews (three were receiving breast milk alone, five were receiving breast and formula milk, and 18 were completely weaned). The kappa index was 0.81, reflecting good concordance. Of the six mothers whose information was discordant, five overestimated and one underestimated the intake of breast milk. With this adjustment the odds ratio for breast and formula milk increased from 3.5to 5.6 and that for formula milk decreased from 9.9 to 6.9 (table 5). Therefore, recall bias may have reduced the estimate of risk for children receiving both breast and formula milk and increased the risk for infants who had been completely weaned. However, our main conclusions remain unchanged. Berkson paradox was controlled for during adjusted analysis,19 and limitation related to diagnostic criteria was reduced by using referees.12

Table 5

Simulation to assess effect of misclassification on odds ratios for being admitted for pneumonia for all children

View this table:

Previous studies

Recent publications have emphasised the need for using standard definitions of feeding patterns to allow comparison between studies.20 In our sample few infants were exclusively breast fed—20% in the first month and 1.6% at three months—because formula milk and herbal teas are widely used.21 The low rate of exclusive breast feeding precludes the use of such infants as the baseline category with the lowest expected risk. We therefore used three different variables (type of milk consumed, intake of fluid supplements, and intake of solid and semisolid supplements) to characterise feeding patterns. With this approach the dose-response effect of the type of milk consumed could be assessed—most studies treat breast feeding as a dichotomous variable 4 18 —and the effects of milk, fluids, and other foods could be separated.

Several studies from less developed countries show that the risk of acquiring an acute lower respiratory infection or pneumonia is 1.5-4 times greater among infants who are not breast fed. 4 18 2227 In our study the risk of admission for pneumonia was 17 times greater among infants who were not being breast fed. Even for children who received both maternal and formula milk, the risk was about four times greater than that for children who received breast milk alone. This marked dose-response effect, along with the biological plausibility of a link between breast feeding and pneumonia, is strongly supportive of a causal association. 4 18 22

A Peruvian study investigated whether the protection of breast feeding against respiratory infections changed with age,24 but it was controlled for few confounding factors. In Brazil the interaction observed was not significant.22 Further research from Brazil,23 Argentina,25 India,26 and China27 did not report interactions between age and breast feeding. In our study the protective effect of breast milk was markedly stronger among young infants than at later ages. This finding is biologically plausible since the immature immune system of young infants is likely to render as even more important the protection afforded by breast milk.3


The relative risks of pneumonia associated with the introduction of supplementary foods also varied markedly with age. To our knowledge, this interaction had not been previously described in the literature.

This study shows that breast feeding protects infants against pneumonia and that this proctection varies considerably according to infant age. These findings reinforce the need for targeting breastfeeding promotion efforts at the mothers of very young infants and for recommending the timely introduction of supplementary foods.


We particularly thank Alexander M Walker, Department of Epidemiology, Harvard School of Public Health, and Saul S Morris, Department of Epidemiology and Population Health, London School of Hygiene and Tropical Medicine for their critical reading of the manuscript and their generous comments. We thank the three independent referees who diagnosed pneumonia from hospital records: Drs Elaine P Albernaz, Luciani M Oliveira, and Ricardo Halpern.

Contributors: All authors participated in the study proposal. JAC collected and analysed the data and wrote the paper. CGV helped in the data analysis and in editing the paper. FCB, ISS, and JAF discussed main ideas and helped to edit the paper. JAC and CGV are guarantors.

Funding: This study was supported by the European Community, World Health Organisation, and Fundaçäo de Amparo a Pesquisa do Rio Grande do Sul (FAPERGS), Brazil.

Competing interests: None declared.