DOI:10.1542/peds.108.4.

e67
2001;108;e67 Pediatrics
and Stan Becker
Shams Arifeen, Robert E. Black, Gretchen Antelman, Abdullah Baqui, Laura Caulfield
Deaths Among Infants in Dhaka Slums
Exclusive Breastfeeding Reduces Acute Respiratory Infection and Diarrhea
http://www.pediatrics.org/cgi/content/full/108/4/e67
located on the World Wide Web at:
The online version of this article, along with updated information and services, is
reserved. Print ISSN: 0031-4005. Online ISSN: 1098-4275.
Village, Illinois, 60007. Copyright 2001 by the American Academy of Pediatrics. All rights
trademarked by the American Academy of Pediatrics, 141 Northwest Point Boulevard, Elk Grove
and publication, it has been published continuously since 1948. PEDIATRICS is owned, published,
PEDIATRICS is the official journal of the American Academy of Pediatrics. A monthly
. Provided by Indonesia:AAP Sponsored on January 31, 2011 www.pediatrics.org Downloaded from
Exclusive Breastfeeding Reduces Acute Respiratory Infection and
Diarrhea Deaths Among Infants in Dhaka Slums
Shams Arifeen, MBBS, DrPH*; Robert E. Black, MD, MPH; Gretchen Antelman, ScD;
Abdullah Baqui, MBBS, DrPH*; Laura Caulfield, PhD; and Stan Becker, PhD
ABSTRACT. Objectives. To describe breastfeeding
practices and investigate the influence of exclusive
breastfeeding in early infancy on the risk of infant
deaths, especially those attributable to respiratory infec-
tions (ARI) and diarrhea.
Methods. A prospective observational study was con-
ducted on a birth cohort of 1677 infants who were born in
slum areas of Dhaka in Bangladesh and followed from
birth to 12 months of age. After enrollment at birth, the
infants were visited 5 more times by 12 months of age.
Verbal autopsy, based on a structured questionnaire, was
used to assign a cause to the 180 reported deaths. Propor-
tional hazards regression models were used to estimate
the effect of breastfeeding practices, introduced as a
time-varying variable, after accounting for other vari-
ables, including birth weight. Overall neonatal, postneo-
natal and infant mortality, and mortality attributable to
ARI and diarrhea were measured.
Results. The proportion of infants who were breast-
fed exclusively was only 6% at enrollment, increasing to
53% at 1 month and then gradually declining to 5% at 6
months of age. Predominant breastfeeding declined from
66% at enrollment to 4% at 12 months of age. Very few
infants were not breastfed, whereas the proportion of
partially breastfed infants increased with age. Breast-
feeding practices did not differ between low and normal
birth weight infants at any age. The overall infant mor-
tality rate was 114 deaths per 1000 live births. Compared
with exclusive breastfeeding in the first few months of
life, partial or no breastfeeding was associated with a
2.23-fold higher risk of infant deaths resulting from all
causes and 2.40- and 3.94-fold higher risk of deaths at-
tributable to ARI and diarrhea, respectively.
Conclusion. The important role of appropriate breast-
feeding practices in the survival of infants is clear from
this analysis. The reduction of ARI deaths underscores
the broad-based beneficial effect of exclusive breastfeed-
ing in prevention of infectious diseases beyond its role in
reducing exposure to contaminated food, which may
have contributed to the strong protection against diar-
rhea deaths. Pediatrics 2001;108(4). URL: http://www.
pediatrics.org/cgi/content/full/108/4/e67; infant mortality,
birth weight, breastfeeding, diarrhea, ARI, Bangladesh.
ABBREVIATIONS. ARI, acute respiratory infection; CI, confi-
dence interval.
T
here is ample evidence of a positive influence
of breastfeeding, especially exclusive breast-
feeding, on the survival of the child
15
but less
information regarding effects on cause-specific mor-
tality, such as that attributable to acute respiratory
infections (ARI) and diarrhea. Yoon et al
6
reported a
higher risk of diarrhea mortality associated with not
breastfeeding, whereas the risk for death attributable
to ARI, although higher, was not statistically signif-
icant. A similar pattern also has been reported from
other studies.
4,7
In a recent meta-analysis of data
from 6 developing countries, breastfeeding provided
a greater degree of protection against diarrhea
deaths than against deaths attributable to ARI in the
first 6 months of life, whereas the level of protection
was similar for infants who were 6 to 11 months of
age.
8
An important limitation of many of these stud-
ies is that they did not use the currently recom-
mended classification of breastfeeding, ie, exclusive,
predominant (almost exclusive), and partial,
9
thus
not permitting the protection afforded by each to be
assessed.
It also is possible that the positive effect of breast-
feeding on survival found in other studies may have
been overestimated as a result of confounding. For
example, infants with lower birth weight may be less
likely to breastfeed or may breastfeed for shorter
durations.
10
Because birth weight is a very important
determinant of infant survival in both developed and
developing countries,
1115
failure to adjust for it or
other such factors could explain the observed posi-
tive effect of breastfeeding on child survival. The
objectives of the current study were to explore the
relationships between birth weight and breastfeed-
ing and to assess the effect of exclusive breastfeeding
in early infancy on the risk of infant death, especially
that attributable to ARI and diarrhea.
METHODS
Study Population and Sample
From November 1993 to June 1995, singleton births to a sample
cohort of women who were residents of slum areas in 5 subdis-
tricts of Dhaka City, Bangladesh, were enrolled. Each infants
weight was measured with SECA beam scales (Hamburg, Ger-
many) accurate to 10 g, and information was collected on feeding
practices, illnesses since birth, and the mothers perception of the
size of the newborn infant. Information on household and parental
From the *International Centre for Diarrhoeal Disease Research, Bang-
ladesh, Dhaka, Bangladesh; and Departments of International Health and
Population and Family Health Sciences, Johns Hopkins University School
of Hygiene and Public Health, Baltimore, Maryland.
Received for publication Oct 2, 2000; accepted May 21, 2001.
Reprint requests to (R.E.B.) Department of International Health, Johns Hop-
kins University School of Hygiene and Public Health, 615 N Wolfe St,
Baltimore, MD 21205. E-mail: rblack@jhsph.edu
PEDIATRICS (ISSN 0031 4005). Copyright 2001 by the American Acad-
emy of Pediatrics.
http://www.pediatrics.org/cgi/content/full/108/4/e67 PEDIATRICS Vol. 108 No. 4 October 2001 1 of 8
. Provided by Indonesia:AAP Sponsored on January 31, 2011 www.pediatrics.org Downloaded from
characteristics and the mothers fertility experience had been col-
lected before birth.
Infants were visited at home at approximately 1, 3, 6, 9, and 12
months of age, and structured questionnaires were used to collect
information on infant feeding and morbidity. For each infant who
died during the study period, information for assigning a cause of
death was collected through a verbal autopsy questionnaire. This
instrument was administered to the infants caregiver (usually the
mother) at the next follow-up visit. The verbal autopsy question-
naire was based on questionnaires used elsewhere in Bang-
ladesh
16,17
and was pretested before use.
Dependent Variables
Overall mortality and mortality attributable to ARI and diar-
rhea were chosen as the dependent variables in this analysis. To
assign cause of death, we used an algorithm that consisted of 4
steps. In step 1, all deaths in the first 3 days of life (02 days) were
classified as early neonatal because the verbal autopsy instru-
ment was inadequate for assigning causes in these very young
infants. In step 2, deaths that were attributable to obvious injuries
were classified as attributable to injury. In step 3, neonatal
tetanus was assigned the cause of death for deaths between 3 and
13 days that manifested convulsions before death, and either the
infant cried normally after birth and stopped crying in the final
illness or the infant suckled normally after birth and stopped
suckling in the final illness. In step 4, deaths were assigned as
either ARI or diarrhea or both as cause of death if they met the
definitions below. Finally, in step 5, deaths that did not meet these
definitions were classified according to criteria for possible causes
of death (as described in Baqui et al
18
) and are grouped under
other causes.
For an ARI death, 2 conditions had to be met: 1) there was
either cough in the 2 weeks before death, starting at least 3 days
before death and lasting until the day before death, or difficulty
breathing in the 2 weeks before death, starting at least 1 day before
death and lasting until the day of death; and 2) at least 3 of the
following symptoms were present before death: cough (as defined
above), difficulty breathing, noise during breathing, nasal flaring,
blue lips/skin, chest in-drawing, rapid breathing, not feeding well
or not able to drink, abnormally sleepy or difficult to wake. For a
diarrhea death, there must have been frequent loose or liquid
stools starting 1 to 13 days before death and lasting until death,
with peak stool frequency of 3 or more and with at least 2 of the
following symptoms present: weakness, thirst, dry mouth, loose
skin, depressed fontanelle, sunken eyes, and no or little urine; or
only frequent loose or liquid stools reported in the 2 weeks before
death starting 14 or more days before death and lasting until
death.
Explanatory Variables
Explanatory variables used in this analysis included 2 key
variables of interestbirth weight and breastfeeding statusand
measured confounding variables. Birth weight was estimated
from weights at enrollment measured as late as 13 days after birth
(median: 2 days). Postnatal change in body weight was investi-
gated in a subsample of 99 newborns with daily repeat weights for
14 days after birth.
19
This analysis revealed very little change in
body weight in the first 72 hours. Thus, enrollment weight mea-
sured within 72 hours of birth was used as birth weight, which
was the case in approximately 75% of the infants. In the remaining
infants, who were enrolled 3 to 13 days after birth, birth weight
was estimated from enrollment weights using a regression model
developed from the subsample (birth weight 2566.0 58.5
age 35.1 age
2
4.1 age
3
4.2 [age 3]
3
, where age is in
days and age 3 0 if age 3 days). Birth weight was used both
as a continuous and a categorical (2500 g or 2500 g) variable.
Birth weight was not known for 23 newborns who had died
before enrollment. Mothers were asked whether the size of their
infant was larger than usual, usual, smaller than usual, or very
small. These categories correlated well with measured birth
weight. For each of the 4 subjective size categories, the mean birth
weight and standard deviation were estimated. Birth weights then
were imputed for the 23 newborns who had already died before
enrollment and who thus had missing weights, ie, their birth
weights were randomly generated from distributions of birth
weights based on the mean and half the standard deviation that
corresponded to the same subjective size among infants who were
enrolled alive.
Information on infant feeding practices in the 7 days preceding
each visit resulted in the classification of the infant into exclusively
breastfed (breast milk and nothing else), predominantly breastfed
(breast milk with water, sugar water, honey, or other nonmilk
liquid), partially breastfed (breast milk with animal/powdered/
condensed milk, cereals, eggs, vegetables, fruits, lentils, or meat/
fish), and not breastfed. Variables were created only from infor-
mation from the months 0, 1, and 3 visits. Because the number of
infants who were not breastfed at these visits was small, they were
grouped with the partially breastfed in the analysis. These were
used as time-varying variables in the proportional hazards regres-
sion models to assess the effect of feeding practices in the first few
months of life on infant mortality. For 157 (9%) infants, there was
missing feeding information at the month 1 and/or 3 visit.
The measured confounding variables included in the analysis
were proximate determinants of growth, such as childs age and
sex, and parental and household variables that act either through
the proximate determinants or by other mechanisms, some of
which are not measurable. The parental and household variables
(mothers age, education, height, place of birth and parity, previ-
ous child death of the mother; fathers education; and household
religion, monthly income, and economic status indicator) were
included in the regression models to account for the factors that
were not measurable. Many of these variables are associated with
growth or mortality risk in children.
19,20
The study was approved by the ethical review committees of
International Centre for Diarrhoeal Disease Research, Bangladesh,
and the Johns Hopkins School of Public Health. Informed verbal
consent was obtained from at least 1 parent/caregiver of the
infant. Sick infants were referred to an appropriate facility. Parents
were encouraged either to seek early care from a health facility or
to obtain assistance from 1 of our 3 field offices to facilitate
referral. A study physician was assigned to the Dhaka Shishu
(Childrens) Hospital to ensure that enrolled infants who were
admitted for illness received appropriate care.
Analysis
Total and cause-specific mortality rates by age were calculated
for selected sociodemographic, economic, and maternal endow-
ment categories. To account for differential follow-up time,
Kaplan-Meier estimates of the survival functions
21
and person-
time calculations were used. Child-years were calculated for neo-
natal (028 days) and postneonatal age groups. Exact times of
follow-up for deaths were used in the calculation of child-years;
however, infants who out-migrated (14%) did so between sched-
uled visits with unknown exit times. For the out-migrated infants,
the time from birth to the last visit when the infant was present
and a time period equal to half the interval between the last
scheduled visits was added to get the person-time. The latter was
1.5 months (0.125 years) for out-migration after the month 3 visit,
1 month (0.083 years) for out-migration between the month 1 and
3 visits, and 0.5 month (0.042 years) for out-migration before the
month 1 visit. All follow-up was truncated at 12 months (365.25
days).
Mortality rate ratios were estimated comparing mortality rates
between different levels of the selected variables, and 95% confi-
dence intervals (CI) were calculated by Miettinens test-based
method.
22,23
The relationship between birth weight and breast-
feeding status was explored using proportions and means. Tests of
significance for differences were performed using
2
tests for
proportions and t tests for means.
Cox proportional hazards regression analysis was performed to
estimate the effect of the key explanatory variables on total and
cause-specific mortality after accounting for the effect of other
confounding variables.
24
Hazard ratios were estimated as the
exponential of the regression coefficients, and 95% CI for the
hazard ratios were calculated.
25
Models were built in 2 stages. In the first, only the confounding
variables were incorporated in 7 different models: for 3 event
types (all deaths, ARI deaths, diarrhea deaths) and 3 age periods
(neonatal, postneonatal, infancy). Two models for neonatal cause-
specific deaths were not constructed as there were only 7 ARI and
4 diarrhea deaths. Deaths attributed to both ARI and diarrhea
were included in both ARI and diarrhea models. In the neonatal
models, follow-up was truncated at 28 days, whereas in the post-
2 of 8 BREASTFEEDING AND INFANT SURVIVAL IN DHAKA SLUMS
. Provided by Indonesia:AAP Sponsored on January 31, 2011 www.pediatrics.org Downloaded from
neonatal models, only those who survived the first 28 days were
included and age was adjusted to start at 0 from age 28 days
(adjusted postneonatal age age 28). Deaths at under 72 hours
of age were not included as events to avoid reverse causality
because the illness preceding these early deaths could have influ-
enced feeding practices before death. Only variables that were
significant at the 0.10 level were retained in the final models. Sets
of dummy variables were not separated, eg, both of the dummy
variables for household income were retained although only 1 of
them met the criteria. Additional variables were dropped from
some of the models for lack of events (deaths) in the reference
categories of those variables. Because of missing values for some
confounding variables, 33 infants were excluded from the final
models.
In the second stage, the 2 key explanatory variables of interest
were added to the base models from stage 1. For events before
the month 1 visit, feeding status at enrollment (month 0) contrib-
uted to the breastfeeding variables in the model. For events be-
tween months 1 and 3 visits, feeding status at the month 1 visit
was the source, whereas the breastfeeding status at the month 3
visit was used for all events that occurred later.
We estimated expected declines in infant mortality as a result of
improvements in breastfeeding practices. The average of the prev-
alence of different breastfeeding categories at the month 1 and 3
visits was taken as a proxy measure of breastfeeding practices in
the first 4 months of life. Mortality rates for each breastfeeding
category then were estimated on the basis of this prevalence
estimate, the overall infant mortality rates from the sample, and
the hazard ratios from the regression. Assuming that these mor-
tality rates and all other factors remain constant, infant mortality
rates then were reestimated for different prevalence levels of
breastfeeding practices. Changes in these prevalence levels as-
sumed shifts from both predominant and partial no breastfeed-
ing to exclusive breastfeeding, and from partial no to predom-
inant breastfeeding.
All analyses were performed with the SAS System software
(version 6.12; SAS Institute, Cary, NC). The procedure PROC
PHREG was used for proportional hazards regression models. The
Corel Quattro Pro (version 7; Ottawa, Canada) spreadsheet pro-
gram was used for estimating expected mortality reductions.
RESULTS
General and Baseline Characteristics of the Sample
The mean age of the mothers of the newborns was
approximately 25 years; almost three quarters did
not have any formal education, and 95% were Mus-
lim. A little more than two thirds of the mothers
were born outside Dhaka, and 23% were shorter than
145 cm. For 26% of the mothers, this was their first
infant and for another 23% the second, and approx-
imately 31% of the mothers had had a previous child
death. Almost 60% of the fathers did not have any
education. The mean monthly household income of
the sample was approximately US $75, and nearly
40% of the households did not have any of the fol-
lowing: television, radio, bicycle, rickshaw, cup-
board, table, and watch/clock. Almost half of the
newborns (47%) were of low birth weight (2500 g),
and mean estimated birth weight was 2516 g.
Mortality and Cause of Death
A total of 180 deaths were identified in the study
cohort by 12 months of age. This translates into a
crude rate of 107 infant deaths per 1000 live births.
Using the complement of the Kaplan-Meier estimate
of the survival function, the overall infant mortality
risk was 114 deaths per 1000 live births. This is a
better estimate as it takes into account losses to fol-
low-up. Exactly one third of these deaths (60) oc-
curred in neonates (028 days), another 99 deaths
(55%) were in the 1- to 5-month age range, and only
21 deaths (12%) happened in the latter half of infancy
(Table 1). Deaths attributable to early neonatal causes
accounted for almost half of the neonatal deaths, 9
were attributable to tetanus (15%), and 6 (10%) could
be attributed to ARI. Only 3 of the neonatal deaths
were attributable to diarrhea, and 1 was attributable
to ARI and diarrhea. During the first part of the
postneonatal period (29 days5 months), ARI was
responsible for one quarter of the deaths, whereas
diarrhea was the cause of death in 20% of the cases.
In 7 cases, both diarrhea and ARI were identified as
causes of death. The pattern was similar in the sec-
ond half of the postneonatal period (611 months),
with ARI being responsible for a greater proportion
of the deaths.
Bivariate Relationships
Except for marginally significant differences at en-
rollment (0 months), low and normal birth weight
infants differ very little with regard to how they were
fed throughout infancy (Table 2). Because most new-
borns were given water, sugar water, honey, or other
liquids for a few days after birth, exclusive breast-
feeding was much less prevalent at this age (overall:
TABLE 1. Distribution of Deaths in the Sample by Cause and Age at Death
Cause of Death All Deaths Age Group
Neonatal
(028 Days)
Postneonatal
(29 Days11 Months)
Early neonatal 27 (15%) 27 (45%)
Injury 4 (2%) 1 (2%) 2 (1%)
Neonatal tetanus 9 (5%) 9 (15%)
Only ARI 39 (22%) 6 (10%) 33 (25%)
Only diarrhea 26 (14%) 3 (5%) 23 (20%)
Both ARI and diarrhea 10 (6%) 1 (2%) 9 (7%)
Other causes
Possible ARI 11 (6%) 5 (8%) 6 (5%)
Possible diarrhea 8 (4%) 0 (0%) 8 (7%)
Possible ARI and diarrhea 5 (3%) 0 (0%) 5 (4%)
Possible malnutrition 2 (1%) 0 (0%) 2 (2%)
Possible malnutrition and ARI 5 (3%) 1 (2%) 4 (3%)
Possible malnutrition and diarrhea 1 (1%) 0 (0%) 1 (1%)
Possible malnutrition, ARI, and diarrhea 5 (3%) 0 (0%) 5 (4%)
Not determined 28 (16%) 7 (12%) 21 (18%)
Total 180 (100%) 60 (100%) 120 (100%)
http://www.pediatrics.org/cgi/content/full/108/4/e67 3 of 8
. Provided by Indonesia:AAP Sponsored on January 31, 2011 www.pediatrics.org Downloaded from
6%). More than half of the infants were breastfed
exclusively at age 1 month, declining to approxi-
mately one quarter at 3 months. After 6 months of
age, very few infants were breastfed exclusively, but
even at 12 months of age, approximately 4% of the
infants were still receiving breast milk with no other
caloric food (predominant). Because very few infants
were not breastfed at any age (maximum: 7% at 12
months of age), their data were combined with those
from the partially breastfed infants.
Neonates who were born to more educated, taller,
or wealthier mothers were less likely to die. If the
infants survived beyond the first month of life, only
maternal education and household income plus
household economic status were associated with re-
duced risk of death (Table 3). The associations be-
tween all other confounding variables and mortality
were statistically nonsignificant. Low birth weight
was a strong determinant of neonatal and postneo-
natal mortality.
We observe a similar trend in the relationship
between the percentage of children who died and
breastfeeding practices in 3 different age periods (Ta-
ble 4). This association is the strongest and the most
obvious in the period between the month 1 and 3
visits, when 7% of the infants who were partially or
not breastfed at month 1 died in the following 2
months compared with approximately 2% of those
who were breastfed exclusively.
Proportional Hazards Regression
Infants who were either partially or not breastfed
had a higher risk of postneonatal death than infants
who were breastfed exclusively for the first 4 months
of life (Table 5). The risk also was slightly greater in
partially or not breastfed neonates but was not sig-
nificant. For infancy, partially or not breastfed in-
fants had 2.23 times the mortality risk of infants who
were breastfed exclusively during the first 4 months.
The hazard ratio estimates for ARI and diarrhea
deaths were 2.40 and 3.94 among partially or not
breastfed infants, compared with exclusively breast-
fed infants. These effects also were seen in the post-
neonatal period. There were too few ARI or diarrhea
TABLE 2. Percentage Distribution of Enrolled Infants by Age, Current Breastfeeding Practices,* and Birth Weight
Age (mo) Birth Weight Category and Breastfeeding Status
Normal Birth Weight (2500 g) Low Birth Weight (2500 g)
Exclusive Predominant Partial
None
Total Exclusive Predominant Partial
None
Total
0 (15 d) 7.5 64.4 28.1 100% 4.7 66.2 29.1 100%
1 (15 d) 53.7 14.8 31.5 100% 52.4 13.9 33.7 100%
3 (15 d) 24.8 15.0 60.2 100% 26.5 12.8 60.7 100%
6 (15 d) 4.8 11.6 83.6 100% 4.8 10.0 85.2 100%
9 (15 d) 1.3 5.7 93.0 100% 0.2 5.2 94.6 100%
12 (15 d) 0.0 4.6 95.4 100% 0.0 3.4 96.6 100%
* Based on breastfeeding practices in the 7 days preceding each visit.
There were no significant differences between normal birth weight and low birth weight infants at any age (at age 0, P .06).
TABLE 3. Effect of Selected Sociodemographic, Economic, and Maternal Endowment Character-
istics on Mortality by Age Group (n 1677)
Characteristic Category Age Group and Rate Ratio
Neonatal Period Postneonatal Period
RR 95% CI RR 95% CI
Birth weight
LBW (2500 g) 2.35 1.403.96 2.21 1.543.17
NBW 1.00 1.00
Maternal schooling
Any 0.46 0.220.96 0.52 0.320.86
None 1.00 1.00
Monthly income
US $60 0.77 0.441.35 0.54 0.350.83
US $4059 0.46 0.230.91 0.71 0.461.10
US $039 1.00 1.00
Household economic status*
High 0.54 0.291.03 0.51 0.330.80
Medium 0.61 0.331.12 0.59 0.380.91
Low 1.00 1.00
Mothers height
150.0168.0 cm 0.77 0.421.40 0.72 0.461.12
145.0149.9 cm 0.51 0.260.98 0.70 0.451.10
126.3144.9 cm 1.00 1.00
RR indicates person-timebased rate ratio; LBW, low birth weight; NBW, normal birth weight.
* A relative indicator of household economic status in a population in which almost all households
live below poverty level; high, household has television, radio, or bicycle/rickshaw; medium, house-
hold has none of the above but has cupboard, table, or watch/clock; low, household has none of the
above.
4 of 8 BREASTFEEDING AND INFANT SURVIVAL IN DHAKA SLUMS
. Provided by Indonesia:AAP Sponsored on January 31, 2011 www.pediatrics.org Downloaded from
deaths in neonates to permit an analysis. Predomi-
nantly breastfed infants did not seem to differ from
those who were given breast milk exclusively, except
for having a higher, though nonsignificant, risk for
diarrheal deaths.
Estimating Expected Reductions in Infant Mortality
The percentage reductions in infant mortality rates
to be expected as a result of breastfeeding promotion
activities at varying levels of effectiveness in a Ban-
gladesh urban slum setting were estimated (Table 6).
The current prevalence of different breastfeeding
practices in the first 4 months of life is based on a
proxy measure (the average of the prevalences at 1
and 3 months of age shown in Table 2). The overall
infant mortality rate of 114 per 1000 live births and
the hazard ratios associated with breastfeeding cate-
gories in the infant models for all deaths (Table 5)
contributed to the estimation. Increase in the preva-
lence of exclusive breastfeeding from current levels
to approximately 78% (twice the current level)
should result in the reduction of infant mortality by
almost one third.
DISCUSSION
This study confirmed the importance of breast-
feeding for infant survival. The study also examined
the relationship between breastfeeding and cause-
specific infant mortality in a community-based set-
ting in a developing country. Approximately 42% of
infant deaths in this population could be attributed
to 2 causes: diarrhea and ARI. This proportion in-
creases to 61% when possible causes are included.
The proportion of infants who died as a result of
these 2 causes is comparable to other studies from
Bangladesh, which used a similar verbal autopsy
instrument.
18
We observed a strong positive effect of exclusive
breastfeeding in the first months of life on infant
survival. Because breastfeeding was modeled as a
time-varying variable, the feeding status at the
month 3 visit was included as the risk factor for all
deaths after that visit. For deaths before the month 3
visit, the feeding status from the immediately pre-
ceding visit was incorporated in the model. This
approach guards against reverse causality because
feeding practices often change during serious illness,
such as an illness preceding death.
Similar to findings from previous studies, the risk
ratio estimates associated with partial or not breast-
feeding were higher for diarrhea deaths than for ARI
deaths, although the CI overlapped. The highest risk
group for diarrhea deaths included infants who were
either not breastfed or who received, in addition to
breast milk, other energy-containing foods at an age
when they were not necessary. Contaminated com-
plementary foods are primary sources of gastrointes-
tinal pathogens
26
and are the most likely explanation
for the observed association. The higher, though
nonsignificant, risk of diarrhea deaths among pre-
dominantly breastfed infants lends support to this
hypothesis because, in addition to breast milk, these
infants were, by definition, receiving noncaloric
drinks, which were likely to be contaminated. Previ-
ous studies also have reported a protective effect
of breastfeeding on diarrhea morbidity and mortali-
ty.
68,2729
There was approximately a 2.5-fold increase in the
risk of ARI deaths when infants received energy-
containing food in addition to breast milk or were
not breastfed in early infancy. Although there is con-
siderable evidence of an increased risk of respiratory
infections associated with not breastfeeding from
both developed and developing countries,
3032
there
have been few reports of increased risk of ARI deaths
among nonbreastfed infants.
7
The hazard ratio for all
deaths also was increased among infants who were
not exclusively or predominantly breastfed. The
most likely explanation for the observed effect on
ARI and all-cause deaths is the known anti-infective
and nutritional properties of breast milk.
3337
This
probably also is responsible for part of the effect on
diarrhea deaths.
A component of the observed effect of breastfeed-
ing may have been through its effect on infant nutri-
tional status. In an earlier analysis of these data, we
found a strong positive relationship between exclu-
sive breastfeeding and infant growth.
18
Additional
evidence in support of this hypothesis was found
when we incorporated a time-varying infant nutri-
tional status variable in the models (data not shown).
In those models, the hazard ratio estimates for the
breastfeeding variable, though still significant and in
the same direction, was closer to the null in compar-
ison with the estimates in Table 5.
The small sample size and the consequent low
power to detect a difference prevents us from reach-
ing a firm conclusion regarding the risk from pre-
dominant compared with exclusive breastfeeding.
We did see an increased risk of deaths attributable to
diarrhea among infants who were predominantly
breastfed, but the CI overlapped 1. The literature on
the effect of predominant breastfeeding on mortality
is very limited. This is attributable mainly to the fact
that predominant breastfeeding, ie, the addition of
water, sugar water, or other nonmilk drinks, seldom
is treated as a separate category. It is either combined
with partial breastfeeding or not distinguished from
TABLE 4. Relationship of Breastfeeding Practices With Neo-
natal and Postneonatal Deaths
n Deaths %
Neonatal period*
Exclusive breastfeeding 104 2 1.9
Predominant breastfeeding 1094 19 1.7
Partial breastfeeding or not breastfeeding 479 12 2.5
Postneonatal period
Period from month 1 to month 3 visits
Exclusive breastfeeding 796 14 1.8
Predominant breastfeeding 260 6 2.3
Partial breastfeeding or not breastfeeding 516 36 7.0
Period from month 3 to month 12 visits
Exclusive breastfeeding 424 15 3.5
Predominant breastfeeding 232 6 2.6
Partial breastfeeding or not breastfeeding 916 42 4.6
* Excluding deaths 3 days of age.
Feeding practice at enrollment.
Feeding practice at month 1 visit.
Feeding practice at month 3 visit.
http://www.pediatrics.org/cgi/content/full/108/4/e67 5 of 8
. Provided by Indonesia:AAP Sponsored on January 31, 2011 www.pediatrics.org Downloaded from
exclusive breastfeeding. Studies have shown that in-
take of water, tea, and other nonmilk drinks in ad-
dition to breast milk increases the risk of diar-
rhea.
27,29,38,39
The data presented do not suggest that birth
weight is an important influence on the breastfeed-
inginfant mortality relationship in this population.
Contrary to reports from other populations,
10,40
birth
weight did not seem to influence breastfeeding sig-
nificantly at any age. One possible explanation is that
because low birth weight is so common, small infants
are accepted as the norm and thus size does not
influence feeding decisions. Unlike many other pop-
ulations, breastfeeding until 12 months is a norm in
this poor Bangladeshi community. Furthermore, be-
cause most deliveries take place at home, low birth
weight infants are not subject to medical interven-
tions that may change their breastfeeding practices.
Unlike most other studies, in which comparisons
usually are made between breastfed and nonbreast-
fed children or the effect of breastfeeding duration is
examined, our classification of breastfeeding status
corresponds to the World Health Organizations rec-
ommended definitions and provides evidence on the
benefits of currently recommended breastfeeding
practices.
41
A recent meta-analysis provided con-
vincing evidence of the protection provided by any
breastfeeding against all infant deaths and deaths
attributable to diarrhea and ARI.
8
This study, con-
ducted in a population in which breastfeeding is
nearly universal, indicates that exclusive breastfeed-
ing in the first few months of life imparts additional
protection. These results, therefore, reinforce the sci-
entific basis of the current recommendations for con-
tinuing exclusive breastfeeding until 4 to 6 months of
age.
42
In this study, the risk of neonatal mortality was
approximately one third of infant mortality. This is
unexpectedly low. It is possible that some early neo-
natal deaths were missed in the sample because the
newborns were not enrolled before death; however,
the rates in this study are consistent with findings
from surveillance data from another sample of the
Dhaka slum population. Neonatal and postneonatal
rates of 31 and 68 per 1000 live births, respectively,
were reported from that surveillance.
43
Breastfeed-
ing data used in this analysis were collected infre-
quently and did not capture changes between visits.
It is reassuring that misclassification of feeding prac-
tices for this reason would reduce rather than exag-
gerate the observed relationships.
Findings from this study indicate that although
breastfeeding in this population is nearly universal,
interventions to improve infant feeding practices
could result in a considerable reduction in infant
mortality. It is worthwhile to note that human im-
munodeficiency virus infection still is extremely un-
common in this population and so does not influence
the assessment of the benefits and risks of breastfeed-
ing or efforts to promote exclusive breastfeeding.
TABLE 5. Adjusted Hazard (Risk) Ratios and 95% CI for Breastfeeding Status From the Propor-
tional Hazards (Cox) Regression Models, by Age Group and Cause of Death
Breastfeeding
Status*
Cause of Death
All Deaths ARI Diarrhea
Neonatal period
Predominant 0.86 (0.203.68)
Partialnone 1.17 (0.265.30)
Postneonatal period
Predominant 0.96 (0.491.89) 0.77 (0.212.85) 1.82 (0.496.79)
Partialnone 2.30 (1.483.59) 2.74 (1.275.92) 4.04 (1.5010.87)
Overall infancy
Predominant 1.13 (0.651.97) 0.91 (0.332.56)# 2.22 (0.677.37)**
Partialnone 2.23 (1.453.44) 2.40 (1.145.04) 3.94 (1.4710.57)
All models exclude deaths 3 days of age. Variables in the models: a continuous birth weight variable,
time-varying variable for breastfeeding status, and selected confounding variables.
* Reference category: exclusive breastfeeding mode.
Other variables in models: height and education; income, education, and parity; none; parity;
religion, income, height, education, and parity; # income, height, and education; and ** parity.
TABLE 6. Estimation of Expected Reduction in Infant Mortality After Improvement in Breast-
feeding Practices in the Bangladesh Urban Slum Setting
Improvement in
Breastfeeding
Practice*
Breastfeeding Practice (Prevalence) Infant
Mortality
Per 1000
Live Births
Percentage
Reduction
in Infant
Mortality
Exclusive Predominant Partial
None
Current 39% 14% 47% 114
Some 54% 18% 27% 99 13
Medium 65% 16% 19% 90 21
High 70% 18% 12% 84 26
Very high 78% 17% 5% 77 32
* Proxy measure of breastfeeding practices in first 4 months of life.
Assumes a risk ratio of 2.30 for partialno breastfeeding and 1.20 for predominant breastfeeding
(Table 5).
Average based on Table 2.
6 of 8 BREASTFEEDING AND INFANT SURVIVAL IN DHAKA SLUMS
. Provided by Indonesia:AAP Sponsored on January 31, 2011 www.pediatrics.org Downloaded from
However, if findings from South Africa that exclu-
sive, as opposed to mixed, breastfeeding protects the
infant from human immunodeficiency virus trans-
mission are confirmed, there will be another impor-
tant reason to promote exclusive breastfeeding.
44
Promotion of exclusive breastfeeding in the first 4
to 6 months of life and reduction in the current
practices of giving potentially contaminated drinks
or foods are likely to be beneficial for infant survival
in this population, with reductions expected in both
diarrhea and ARI deaths, as well as in all deaths. The
limited number of infants who were breastfed exclu-
sively to 6 months of age as currently recommended
precluded an assessment of protection in the fifth
and sixth month of life, so the full benefits may be
even greater if these recommendations were fol-
lowed. Although somewhat simplistic, the estimates
in Table 6 provide a measure of the expected benefits
in infant survival as a result of breastfeeding promo-
tion activities. They are consistent with earlier esti-
mates of potential reductions in diarrhea mortality
after increased breastfeeding.
5
Thus, estimates indi-
cate that infant mortality could be reduced by almost
one third if the prevalence of exclusive breastfeeding
in the first 4 months of life could be raised to nearly
80%, with smaller gains with intermediate improve-
ments in breastfeeding practices. Recent studies in
Bangladesh showed that such improvements in
breastfeeding practices can be achieved through
community-based interventions.
45
ACKNOWLEDGMENTS
This research was supported by the Royal Netherlands Gov-
ernment under Agreement No. BD009602 with ICDDR,B, the
United States Agency for International Development (USAID)
under Cooperative Agreement No. 399-0073-A-00-1054-02 with
ICDDR,B, and under the Johns Hopkins Family Health and Child
Survival Cooperative Agreement No. HRN-A-00-96-9006-00, and
the ICDDR,B: Center for Health and Population Research. The
Center is supported by more than 30 countries and agencies that
share its concern for the health and population problems of de-
veloping countries.
S.A., R.B., G.A., and A.B. conceptualized and designed the
study. S.A. and G.A. were responsible for the implementation
of field work. S.A., R.B., L.C., and S.B. designed the analysis. S.A.
drafted the manuscript, which then was revised by all investiga-
tors.
REFERENCES
1. Shahidullah M. Breast-feeding and child survival in Matlab, Bang-
ladesh. J Biosoc Sci. 1994;26:143154
2. de Zoysa I, Rea M, Martines J. Why promote breast-feeding in diarrheal
disease control programmes? Health Policy Plann. 1991;6:371379
3. Habicht JP, DaVanzo J, Butz WP. Does breast-feeding really save lives,
or are potential benefits due to biases? Am J Epidemiol. 1986;123:279290
4. Jason JM, Nieburg P, Marks JS. Mortality and infectious disease asso-
ciated with infant-feeding practices in developing countries. Pediatrics.
1984;74(suppl):702727
5. Feacham RG, Koblinsky MA. Interventions for the control of diarrheal
diseases among young children: promotion of breast-feeding. Bull
World Health Organ. 1984;62:271291
6. Yoon PW, Black RE, Moulton LM, Becker S. Effect of not breastfeeding
on the risk of diarrheal and respiratory mortality in children under 2
years of age in Metro Cebu, The Philippines. Am J Epidemiol. 1997;143:
11421148
7. Victora CG, Smith PG, Vaughan JP, et al. Evidence for protection by
breast-feeding against infant deaths due from infectious diseases in
Brazil. Lancet. 1987;2:319322
8. WHO Collaborative Study Team on the Role of Breastfeeding on the
Prevention of Infant Mortality. Effect of breastfeeding on infant and
child mortality due to infectious diseases in less developed countries: a
pooled analysis. Lancet. 2000;355:451455
9. Labbok M, Krasovec K. Toward consistency in breastfeeding defini-
tions. Stud Fam Plann. 1990;21:226230
10. Barros FC, Victora CG, Vaughan JP, Smith PG. Birth weight and dura-
tion of breast-feeding: are the beneficial effects of human milk being
overestimated? Pediatrics. 1986;78:656661
11. McCormick MC. The contribution of low birth weight to infant mortal-
ity and childhood morbidity. N Engl J Med. 1985;312:8290
12. Kramer MS. Determinants of low birth weight: methodological assess-
ment and meta-analysis. Bull World Health Organ. 1987;65:663737
13. Mata LJ, Urrutia JJ, Kronmal RA, Joplin C. Survival and physical growth
in infancy and early childhood: study of birth weight and gestational
age in a Guatemalan Indian village. Am J Dis Child. 1975;129:561566
14. Victora CG, Barros FC, Vaughan JP, Teixeira AMB. Birthweight and
infant mortality: a longitudinal study of 5914 Brazilian children. Int J
Epidemiol. 1987;16:239245
15. Ghosh S, Ramanujacharyulu TKTS, Hooja V, Madhavan S. Mortality
pattern in an urban birth cohort. Indian J Med Res. 1979;69:616623
16. Kalter HD, Hossain M, Burnham G, et al. Validation of caregiver inter-
views to diagnose common causes of severe neonatal illness. Paediatr
Perinat Epidemiol. 1999;13:99113
17. Osinski P. Child Characteristics, Diarrheal Episode Characteristics, and
Mothers Interpretations of Diarrheal Episodes as Determinants of the Use of
Oral Rehydration Therapy in Rural and Urban Bangladesh. [dissertation]
Baltimore, MD: Johns Hopkins University School of Hygiene and Public
Health; 1991
18. Baqui AH, Black RE, Arifeen SE, Hill K, Mitra SN, Sabir AA. Causes of
childhood deaths in Bangladesh: results of a nation-wide verbal au-
topsy study. Bull World Health Organ. 1998;76:161171
19. Arifeen SE. Birth Weight, Intrauterine Growth Retardation and Prematurity:
A Prospective Study of Infant Growth and Survival in the Slums of Dhaka,
Bangladesh [dissertation]. Baltimore, MD: Johns Hopkins University
School of Hygiene and Public Health; 1997
20. Pelletier DL, Low JW, Johnson FC, Msukwa LA. Child anthropometry
and mortality in Malawi: testing for effect modification by age and
length of follow-up and confounding by socioeconomic factors. J Nutr.
1994;124(suppl):2082S2105S
21. Kaplan EL, Meier P. Nonparametric estimation from incomplete obser-
vations. J Am Stat Assoc. 1958;53:457481
22. Miettinen O. Estimability and estimation in case-referent studies. Am J
Epidemiol. 1976;103:226235
23. Kleinbaum DG, Kupper LL, Morgenstern H. Epidemiologic Research:
Principles and Quantitative Methods. New York, NY: Van Nostrand Re-
inhold Company Inc; 1982
24. Cox DR. Regression models and life tables. J R Stat Soc. 1972;34:187220
25. Lee ET. Statistical Methods for Survival Data Analysis. 2nd ed. New York,
NY: John Wiley & Sons Inc.; 1992
26. Black RE, Brown KH, Becker S, Alim AR, Merson MH. Contamination
of weaning foods and transmission of enterotoxigenic Escherichia coli
diarrhea in children in rural Bangladesh. Trans R Soc Trop Med Hyg.
1982;76:259264
27. Mondal SK, Gupta PG, Gupta DN, et al. Occurrence of diarrheal dis-
eases in relation to infant feeding practices in a rural community in
West Bengal, India. Acta Paediatr. 1996;85:11591162
28. Ashworth A. Nutrition interventions to reduce diarrhoea morbidity and
mortality. Proc Nutr Soc. 1998;57:167174
29. Brown KH, Black RE, Lopez de Romana G, Creed de Kanashiro H.
Infant-feeding practices and their relationship with diarrheal and other
diseases in Huascar (Lima), Peru. Pediatrics. 1989;83:3140
30. Cushing AH, Samet JM, Lambert WE, et al. Breastfeeding reduces risk
of respiratory illness in infants. Am J Epidemiol. 1998;147:863870
31. Forman MR, Graubard BI, Hoffman HJ, Beren B, Harley EE, Bennett P.
The Pima infant feeding study: breast-feeding and respiratory infections
during the first year of life. Int J Epidemiol. 1984;13:447453
32. Cunningham AS, Jelliffe DB, Jellife EFP. Breast-feeding and health in
the 1980s: a global epidemiologic review. J Pediatr. 1991;118:659666
33. Akre J, ed. Infant feeding: the physiological basis. Bull World Health
Organ. 1989;67(suppl):1108
34. Black RE, Lopez de Romana G, Brown KH, Tacket CO. Incidence and
etiology of infantile diarrhea and major routes of transmission in Hua-
rcar, Peru. Am J Epidemiol. 1989;129:785799
35. Clemens JD, Stanton B, Stoll B, Shahid NS, Banu H, Chowdhury AK.
Breastfeeding as a determinant of severity in shigellosis. Evidence for
protection throughout the first three years of life in Bangladeshi chil-
dren. Am J Epidemiol. 1986;123:710720
36. Khin-Maung U, Wai NN, Khin M, Khin MM, Tin U, Thane-Toe. Effect
http://www.pediatrics.org/cgi/content/full/108/4/e67 7 of 8
. Provided by Indonesia:AAP Sponsored on January 31, 2011 www.pediatrics.org Downloaded from
on clinical outcome of breastfeeding during acute diarrhoea. Br Med J.
1985;290:587589
37. Rowland MGM, Goh SGJ, Cole TJ. Impact of infection on the growth of
children from 0 to 2 years in a West African community. Am J Clin Nutr.
1988;47:134138
38. Martines JC, Habicht JP, Ashworth A, Kirkwood BR. Weaning in South-
ern Brazil: is there a weanlings dilemma? J Nutr. 1994;124:11891198
39. Popkin BM, Adair L, Akin JS, Black R, Briscoe J, Flieger W. Breast-
feeding and diarrheal morbidity. Pediatrics. 1990;86:874882
40. Adair L, Popkin BM. Low birth weight reduces the likelihood of breast-
feeding among Filipino infants. J Nutr. 1996;126:103112
41. World Health Organization. Complementary Feeding of Young Children in
Developing Countries: A Review of Current Scientific Knowledge. Geneva,
Switzerland: World Health Organization; 1998 (WHO/NUT/98.1)
42. World Health Organization. The World Health Organizations infant-
feeding recommendation. Wkly Epid Rec. 1995;17:117220
43. Baqui AH, Islam R, Begum N, et al. Urban Panel Survey: Dhaka. Charac-
teristics of Sampled Population, Demographic Events, Fertility Regulation,
and Sources of MCH-FP Services, 1995. Dhaka, Bangladesh: International
Centre for Diarrhoeal Disease Research, Bangladesh; 1995 (ICDDR,B
Scientific Report No. 81)
44. Coutsoudis A, Pillay K, Spooner E, Kuhn L, Coovadia HM, for the South
African Vitamin A Study Group. Influence of infant-feeding patterns on
early mother-to-child transmission of HIV-1 in Durban, South Africa: a
prospective cohort study. Lancet. 1999;354:471476
45. Haider R, Ashworth A, Kabir I, Huttly SR. Effect of community-based
peer counsellors on exclusive breastfeeding practices in Dhaka,
Bangladesh: a randomised controlled trial. Lancet. 2000;356:16431647
8 of 8 BREASTFEEDING AND INFANT SURVIVAL IN DHAKA SLUMS
. Provided by Indonesia:AAP Sponsored on January 31, 2011 www.pediatrics.org Downloaded from
DOI:10.1542/peds.108.4.e67
2001;108;e67 Pediatrics
and Stan Becker
Shams Arifeen, Robert E. Black, Gretchen Antelman, Abdullah Baqui, Laura Caulfield
Deaths Among Infants in Dhaka Slums
Exclusive Breastfeeding Reduces Acute Respiratory Infection and Diarrhea
& Services
Updated Information
http://www.pediatrics.org/cgi/content/full/108/4/e67
including high-resolution figures, can be found at:
References
http://www.pediatrics.org/cgi/content/full/108/4/e67#BIBL
at:
This article cites 33 articles, 15 of which you can access for free
Subspecialty Collections
http://www.pediatrics.org/cgi/collection/infectious_disease
Infectious Disease & Immunity
following collection(s):
This article, along with others on similar topics, appears in the
Permissions & Licensing
http://www.pediatrics.org/misc/Permissions.shtml
tables) or in its entirety can be found online at:
Information about reproducing this article in parts (figures,
Reprints
http://www.pediatrics.org/misc/reprints.shtml
Information about ordering reprints can be found online:
. Provided by Indonesia:AAP Sponsored on January 31, 2011 www.pediatrics.org Downloaded from