Miotti PG, Taha TET, Kumwenda NI, Broadhead R, Mtimavalye LAR, Van der Hoeven L, Chiphangwi JD, Liomba G, Biggar RJ. HIV Transmission Through BreastfeedingA Study in Malawi. JAMA. 1999;282(8):744-749. doi:10.1001/jama.282.8.744
Author Affiliations: Department of Epidemiology, Johns Hopkins School of Hygiene and Public Health, Baltimore, Md (Drs Miotti, Taha, and Kumwenda, and Ms Van der Hoeven); Malawi College of Medicine, Blantyre, Malawi (Drs Broadhead, Mtimavalye, Chiphangwi, and Liomba); and the National Cancer Institute, National Institutes of Health, Bethesda, Md (Dr Biggar). Dr Miotti is currently with the National Institute of Allergy and Infectious Diseases, National Institutes of Health, Bethesda, Md.
Context Understanding the risk of human immunodeficiency virus (HIV) transmission
through breastfeeding is essential for advising HIV-infected mothers and formulating
public health policy recommendations.
Objective To measure the frequency, timing, and risk factors of HIV transmission
through breast milk.
Design Prospective cohort study conducted between 1994 and 1997, with follow-up
of infants through 24 months of age.
Setting Postnatal clinic of tertiary care hospital, Blantyre, Malawi.
Participants A total of 672 infants (HIV-negative at birth) born to HIV-infected
women who had not received antiretroviral drugs during or after pregnancy.
Main Outcome Measure Incidence of HIV in breastfed infants by age and maternal and infant
risk factors for HIV transmission, using proportional hazard models to derive
risk ratios (RRs) and 95% confidence intervals (CIs).
Results Forty-seven children became HIV-infected while breastfeeding but none
after breastfeeding had stopped. The cumulative infection rate while breastfeeding,
from month 1 to the end of months 5, 11, 17, and 23, was 3.5%, 7.0%, 8.9%,
and 10.3%, respectively. Incidence per month was 0.7% during age 1 to 5 months,
0.6% during age 6 to 11 months, and 0.3% during age 12 to 17 months (P=.01 for trend). The only factors significantly associated
with low risk of postnatal HIV transmission in a multivariate model were high
maternal parity (RR, 0.23; 95% CI, 0.09-0.56) and older maternal age (RR,
0.44; 95% CI, 0.23-0.84).
Conclusions Our data suggest that the risk of HIV infection is highest in the early
months of breastfeeding, which should be considered in formulating breastfeeding
Mother-to-child transmission of human immunodeficiency virus (HIV) can
occur in utero, intrapartum, and postnatally.1,2
Postnatal HIV transmission through HIV-contaminated breast milk is of particular
concern in many developing countries, where HIV infection in women is common
and breastfeeding is almost universally practiced. Transmission of HIV through
breast milk has been documented in many studies,3- 12
and HIV has been found in breast milk samples of HIV-infected women.13- 16
Ascertaining the transmission risk of HIV at different times during
the breastfeeding period has become particularly important, because it has
recently been shown that in utero and intrapartum transmission can be decreased
by approximately 50% when short-course, oral antiretroviral therapy is used
during pregnancy through labor.17 In breastfeeding
populations, however, any decrease in in utero and intrapartum transmission
of HIV achieved through such regimens or other methods of prevention will
result in a larger number of infants, who, though uninfected at birth, become
exposed to HIV through breast milk.
In this study, we investigated the risk of HIV transmission through
breastfeeding in an urban setting in Malawi, where HIV prevalence in nursing
women is approximately 30%, and breastfeeding is the recommended method of
infant feeding. A revised statement in 1998 by the Joint United Nations Programme
on HIV/AIDS18 recommended that women be offered
HIV counseling and testing, that they be informed of risks and benefits of
breastfeeding if the mother is HIV-infected, and that they make a decision
that takes into account their individual and family situations. A better understanding
of the level of risk and the timing of infant HIV infection throughout the
breastfeeding period will help to inform women about transmission risks and
to assess policy options about breastfeeding by HIV-infected women.
Between 1994 and 1997, we studied children of HIV-infected women who
had previously participated in a clinical trial of birth canal cleansing to
prevent mother-to-child transmission of HIV. Both the clinical trial and the
follow-up study on breastfeeding risk took place at Queen Elizabeth Central
Hospital, Blantyre, a tertiary care hospital in which approximately 14,000
deliveries occur annually. The follow-up included 37 twin and triplet births.
No mothers in the study received antiretrovirals during pregnancy or postnatally.
The protocol and consent forms were approved by institutional review boards
in Malawi and the United States.
Figure 1 shows the number
of study infants, those who were excluded from the study, and the reasons
for exclusion. Infants found to be HIV-positive by polymerase chain reaction
(PCR) at their first postnatal visit, scheduled at age 6 weeks (n=355), were
excluded from the study, because most of them would have been infected in
utero or perinatally. A small proportion, however, would have been infected
through colostrum or early milk. To be included in the study, breastfed infants
had to be HIV-negative at their first postnatal visit, have breastfeeding
data (n=1012), and have a second follow-up visit (n=672). The interval between
the 2 visits was the observation period during which breastfeeding risk was
assessed. All postnatal HIV infections in infants were considered to be caused
by breastfeeding, because no infant infection was documented after breastfeeding
had stopped. Although all blood donations at this hospital were screened for
HIV, the history of infants' blood transfusions was also reviewed to exclude
possible HIV infection from antibody-negative, virus-infected blood.
Enrolled infants and their mothers were scheduled to attend the study
clinic at 1½, 3, 6, 9, and 12 months after delivery and at 15, 18,
and 24 months in the second year. At each visit, a history of health complaints,
breastfeeding practices, and breast problems (painful swelling, other signs
of infection, cracked or bloody nipples) was obtained. Questions about breast
problems started at the 6-month visit. Physical examination and collection
of biological samples were performed every 6 months. In accordance with World
Health Organization and government of Malawi recommendations at the time of
this study, HIV-infected women, who were told their HIV-infection status,
were not discouraged from breastfeeding. Routine clinical care and, when necessary,
referrals were provided to mothers and infants.
The time of infants' HIV infection was estimated as the midpoint between
the last negative and the first positive PCR result. For infants who did not
seroconvert, the date of the last negative PCR result was used as the date
at which the infant was still uninfected. Infants were censored on the date
when breastfeeding was stopped, as ascertained by the mother's interview.
If breastfeeding was discontinued during the infection interval, the first
positive HIV result was moved backward in time to the date of last breastfeeding.
Because infants were more closely followed in the first few months of life,
the intervals were smaller for younger than for older infants. Thus, midpoint
estimates are more reliable for earlier months of life. For infants who were
weaned but for whom the exact weaning date was not provided, the midpoint
between the date of last known breastfeeding and the date when the infant
was known to have been weaned was used. Observations were truncated at 24
months, because the follow-up data after that time were sparse (16 person-months
of follow-up between 24 and 31 months, with no additional HIV infections).
The time when breastfeeding ended was based on the mother's report.
We examined the duration of breastfeeding for all HIV-infected and uninfected
women to determine if breastfeeding patterns were affected by informing the
women of their HIV status. We also examined transmission risk in uninfected
infants after they had been weaned.
The mothers' HIV infection was established by repeated positive results
from enzyme-linked immunosorbent assay (ELISA) HIV antibody tests performed
on umbilical cord blood. Borderline HIV ELISA results were confirmed by immunoblotting.
Infants' HIV status was determined with DNA PCR using blood collected by heel
stick and adsorbed on filter paper, as previously described.19
Positive and negative test results in Malawian infants had predictive values
of 98% and at least 96%, respectively.19 The
presence of HIV antibody (ELISA) was used to confirm HIV infection in infants
aged 15 months or older, and all new, repeated positive results in infants
were confirmed by immunoblotting.
For analysis, the time at risk for HIV infection was the time under
observation from the infant's first postnatal negative HIV PCR result. Risk
was assessed by Kaplan-Meier method.20 Period
risk was derived from cumulative risk at 6-month intervals, such that the
additional risk in each period was divided by the months during the interval
to provide an average monthly hazard rate. Because 1 criterion for entry into
the study was a negative PCR result after the first month, the first 6 months
of life had no data about risk in the first month of life, and the average
monthly hazard rate is therefore based on only 5 months. Tests of trend assumed
a Poisson distribution. To examine the association between maternal, delivery,
and infant factors and mother-to-child HIV transmission, we used a univariate
proportional hazard model, with left and right truncation of data dependent
on time of entering and leaving the study. The significance of risk factors
was also assessed by a backward stepwise regression model, a multivariate
model that starts with all factors and proceeds by stepwise removal of nonsignificant
Cofactor analysis was expressed as risk ratios (RRs) and confidence
intervals (CIs). Analysis of the time breastfeeding stopped was done by Kaplan-Meier
method,20 with infants censored when breastfeeding
was stopped or when follow-up ended.
Median age of first visit with a negative HIV PCR result was 1.7 months
(25%-75% interquartile range, 1.4-2.1; range, 0.7-16.8 months). Thus, any
postnatal infections did not involve colostrum or very early milk. Infants
uninfected at enrollment were followed up for 7155 person-months (596 person-years)
while breastfeeding. No infant became infected after breastfeeding had stopped
(268 person-months of follow-up), and none of the infants who were found to
be PCR-positive had blood transfusions during the interval of infection. Thus,
all of the new infections were attributable to breastfeeding.
Of the 672 infants with at least 2 follow-up visits (Figure 1), 47 became HIV-infected through breastfeeding. The median
duration of follow-up while breastfeeding was 11.5 months (25%-75% interquartile
range, 4.5-22.0 months). There were no data for the first month, because entry
required a negative PCR result at the first postnatal visit (see "Methods,
Study Population" section). Infants were followed up for a total of 2034 person-months
in the 1- to 5-month interval (21 infections), 2375 person-months in the 6-
to 11-month interval (15 infections), 1995 person-months in the 12- to 17-month
interval (7 infections), and 735 person-months in the 18- to 23-month interval
The postnatal HIV-infection rate is shown in Figure 2. The cumulative risk of infection
for infants continuing to breastfeed after month 1 was 3.5% at the end of
5 months, 7.0% at the end of 11 months, 8.9% at the end of 17 months, and
10.3% at the end of 23 months. The HIV-infection rates per person-month in
the first 2 years of life were 0.7% (months 1-5), 0.6% (months 6-11), 0.3%
(months 12-17), and 0.2% (month 18-23) (Figure
3). This decline in HIV incidence was statistically significant
Table 1 presents
the univariate analysis of the association between infant HIV infection through
breastfeeding and maternal, delivery, and infant factors. Reported maternal
HIV-related illnesses or maternal deaths within 2 years of giving birth were
analyzed, because symptomatic women with poorer immune systems and possibly
higher viral levels may be more likely to transmit HIV infection through breast
milk.13 Compared with asymptomatic mothers,
symptomatic mothers and mothers who died had a similar transmission rate (RR,
1.2; 95% CI, 0.6-2.6). We postulated that older mothers may have been infected
longer and thus may have had more advanced HIV disease. However, the postnatal
transmission RR for women younger than 25 years was not different from that
among older women (0.9; 95% CI, 0.5-1.6). Similarly, women who had several
children and were therefore older may have been HIV-infected longer than women
with fewer children. A lower postnatal transmission rate, however, was found
for higher (≥4) compared with low birth order (<4) infants (RR, 0.4;
95% CI, 0.2-0.9). The higher transmission rate to earlier births was not attributable
to a higher transmission rate to first-born infants, because the greatest
risk was to second-born infants.
The postnatal HIV-infection status
of infants delivered by cesarean birth was analyzed, because we found previously
that in these infants, cesarean delivery was associated with a lower infant
HIV-infection rate than vaginal delivery.21
Thus, infants who avoided infection through a cesarean delivery could have
been at a higher risk of acquiring HIV infection postnatally. The RR for postnatal
transmission was 1.6 (95% CI, 0.8-3.2) in cesarean-delivered infants compared
with vaginally delivered infants.
Breast problems (painful swelling,
infections, and cracked or bloody nipples) reported in the interval in which
infant infection occurred were considered. The HIV transmission rate in women
with and without breast complaints was similar (RR, 0.8; 95% CI, 0.3-2.3).
This assessment only applies to older infants, because the question was asked
at 6 months or later. It is based on a smaller number of infections (n=28)
and may be valid only for breastfeeding after the first 6 months. We studied
the association of infant birth weight, a weak predictor of milk intake, and
postnatal HIV infection, postulating that heavier infants may have had a higher
risk of postnatal infection if they ingested larger volumes of HIV-contaminated
milk. This association was not significant. The RR for HIV infection was lower
in infants weighing 2500 g or more (RR, 0.7; 95% CI, 0.3-1.6) than it was
in smaller infants.
In a backward stepwise regression model, higher
parity (RR, 0.23; 95% CI, 0.09-0.56) and older maternal age (RR, 0.44; 95%
CI, 0.23-0.84) were both significantly protective against postnatal HIV infection
in infants. This model excluded breast problem data, which were only available
for some women. In a separate model, applied to the set of women for whom
data about breast problems were available, the point estimates were generally
similar, but the CI was larger.
These risk factor analyses could
have been biased by late entry into the study, because all infants had to
be HIV PCR–negative after the first month of life to enter the study.
We reexamined data, including every breastfed child who had a postnatal HIV-negative
result, even if they had no further follow-up (n=1012). This approach yielded
similar results in the risk factor analyses, suggesting no major selection
bias among infants who returned for a later follow-up visit.
Among the 672 infants who constituted the follow-up group in this study, the
median time of weaning was 21 months (25%-75% interquartile range, 18-24 months).
Weaning times by 1511 HIV-infected women (median, 22 months [25%-75% interquartile
range, 18-25 months]) and by 3449 uninfected women (median, 22 months [25%-75%
interquartile range, 19-25 months]) were similar, suggesting no bias in the
weaning practices for enrolled infants. Supplemental foods (typically porridge)
were introduced at a mean (SD) time of 4.0 (1.5) months.
In a previous study,22 conducted in a
similar population of women in urban Malawi, we documented a prepartum, peripartum,
and early postnatal HIV transmission rate of 27% among women with vaginal
deliveries. The current study extends those results by documenting the HIV-infection
rate in infants still uninfected at their first postnatal visit. Our study
showed that an uninfected infant, breastfed by an HIV-positive mother for
23 months, had at least a 10.3% risk of becoming infected. This rate of postnatal
infection does not include postnatal transmission in the first month of life,
which could not be reliably distinguished from intrapartum transmission in
this study. The transmission rate in the first month could be substantially
higher than in later months, because it includes feeding with colostrum and
early milk, which are rich in cells.23 Our
trend data showed a higher risk in the 1- to 5-month period than after 5 months,
but we have no data about risk in the first month. Investigators in Brazil24 noted a substantially higher risk among breastfed
infants, even though the average duration of breastfeeding was only about
1 month. A high early transmission rate might be explained by the immaturity
of the infant's immune system and by the large number of HIV-infected cells
present in early milk.23 The overall risk of
postnatal HIV transmission documented in most previous studies ranged from
4% to more than 20%,3,5- 12
with 1 study8 estimating a 32% risk in infants
breastfed for more than 15 months. The interpretation and comparability of
these studies are limited by factors such as small sample size, lack of an
assay capable of reliably detecting infection in early infancy,25
and different ways of defining and calculating postnatal transmission risk.26 At least 1 study27
also found a direct relationship between infant infection risk and breastfeeding
duration. A meta-analysis5 of 5 studies estimated
the HIV transmission risk through breastfeeding to be 14% (95% CI, 7%-22%).
A recent multicenter study28 used
a standardized method to establish rate and time of postnatal transmission
in a large number of children from 4 cohorts in Europe and 4 cohorts in Africa.
Among 902 African infants, 49 postnatal HIV infections were documented, for
an overall incidence of 3.2 per 100 person-years of breastfeeding. This incidence
was less than half the incidence that we observed in Malawi (6.9 per 100 person-years).
A large contribution to the data in the multicenter study came from Kenya,
which had a substantially lower HIV transmission rate compared with the other
3 African studies presented and with our study. In the multicenter study,
the time of HIV infection could be estimated in 20 of the 49 infected infants.
Infants with positive PCR test results in the first few months (mean, 2.5)
of life were excluded from the analysis, as they were in our study. Both studies,
therefore, underestimate the actual postnatal transmission rate. Of the periods
that were analyzed, the largest difference in transmission between the 2 studies
was in the 1- to 5-month period (5.2% in Malawi compared with 0.7% in the
multicenter study). After infants reached 6 months of age, differences in
transmission rates between the 2 studies were less pronounced (6-11 months,
3.4% vs 1.8%; 12-17 months, 2.0% vs 3.8%). Different patterns of weaning and
food supplementation may explain these variations. In our population, most
infants were weaned between 18 and 25 months (median, 22 months), and the
HIV status of the mother did not influence weaning patterns. However, supplemental
foods were introduced well before weaning occurred (median, 4 months) and
the total volume of breast milk ingested, hundreds of milliliters daily,29 might have been influenced by giving supplemental
foods. However, the time of introduction of supplemental foods was not a statistically
significant factor in a time-dependent covariate analysis in our study (P = .2018).
Recommendations for advising HIV-infected
women about breastfeeding require an understanding of the risk factors that
might affect transmission risk.30 Our analysis
of maternal and infant risk factors, however, identified only lower parity
and possibly lower age of the mother as risk factors for HIV transmission
to infants. No significant differences were associated with indicators of
maternal HIV-related disease or early death. Among infant factors, birth weight
was examined as an indicator of larger milk intake, but no significant association
with postnatal HIV infection was found. Our ability to determine statistical
significance in the analysis of risk factors may have been limited by the
small number of postnatal HIV infections. The finding of a higher postnatal
transmission risk in women with lower parity was unexpected, as the women's
younger age would likely be accompanied by lesser virologic and immunologic
compromise. This finding may be consistent with a hypothesis that mothers
who are less experienced with breastfeeding are more likely to have subclinical
mastitis and thereby a higher HIV transmission rate.31
This hypothesis may also be consistent with the higher risk of transmission
that we documented in the early months of breastfeeding. Surprisingly, clinical
mastitis or cracked nipples in mothers were not associated with higher HIV
transmission to the infant. A possible explanation is that women avoid feeding
from a breast that is tender or has overt lesions.
The main limitation
in our study was the lack of data about maternal immunologic and virologic
status. These parameters could influence the rate of HIV transmission through
breastfeeding.13 Active viral replication may
account for the high transmission risk previously documented in mothers who
breastfeed during acute HIV infection.5 Our
study did not investigate mothers who became HIV-infected during lactation.
A second, unavoidable limitation was the inability to determine the transmission
risk through breastfeeding in the first weeks of life. In Kenya, a study currently
in progress has randomized infants into breastfeeding vs bottle-feeding,32 an approach that should provide valuable insights
into the risk from early breastfeeding.
Early weaning has been
as a means to decrease the chance of postnatal HIV transmission. However,
the adverse effects of early weaning on growth, morbidity, and mortality have
been emphasized.33 We found that the hazard
rate of postnatal HIV infection was highest in the first half-year of life,
when breastfeeding is particularly important. Furthermore, HIV transmission
continued for as long as breastfeeding continued. The high level of risk in
the first 6 months of life limits the value of early weaning as a way of reducing
the risk of late postnatal HIV transmission. In the multicenter study,28 weaning at 4 months would have resulted in no infections
and at 6 months in only 3 infections. However, in Malawi, weaning at 4 months
would still have resulted in 19 infections and at 6 months in 21 infections,
which is nearly half of all infections observed in this study. Moreover, both
studies could not detect HIV infection through breastfeeding in the first
several weeks of life. Although some infections would undoubtedly be prevented
by early weaning, there might also be additional illnesses and deaths due
to early weaning. The risk-benefit ratio is thus likely to be widely different
in disparate populations, making it difficult to generalize any weaning recommendation
and its optimal timing.
Several conclusions and possible recommendations
can be drawn from this study. First, the HIV transmission risk due to breastfeeding
was highest in the early months of life, but remained substantial for as long
as an infant continued to breastfeed. Our study's postnatal HIV transmission
rate and other published rates are underestimates, because they do not include
postnatal infections acquired very early in life through breastfeeding. During
early breastfeeding, transmission rates could be substantial.
Second, only lower parity and younger maternal age increased transmission
risk, factors that may be attributable to the degree of maternal experience
with breastfeeding. No obvious factors predicted groups of women who were
clearly more likely to transmit HIV through breastfeeding. Thus, recommendations
could not be specifically directed at high-risk HIV-infected women. The social
implications of recommending that all HIV-infected women avoid breastfeeding
should also be considered, because not breastfeeding may result in social
stigma if it identifies women who are HIV-infected. Perhaps further educational
efforts to instruct young and inexperienced mothers about breastfeeding might
help to lower the likelihood of subclinical mastitis. The role of subclinical
mastitis, however, needs to be further elucidated.
weaning, eg, at 4 or 6 months, is a strategy difficult to generalize. In our
study, it would have prevented only about half of the postnatal infections,
but also may have resulted in additional morbidity and mortality. Breastfeeding
recommendations for HIV-infected women must carefully balance the risk of
HIV transmission with the well-known benefits of breastfeeding (nutritional
excellence, reduced morbidity and mortality, psychological and pregnancy-spacing
benefits). Recommendations may be most usefully made at the level of the individual
mother, because communities in developing countries include women from various
socioeconomic strata who have different access to safe breast milk alternatives.