Relationship between mortality and feeding modality
among children born to HIV-infected mothers in a
research setting: the Kesho Bora Study
Amandine Cournila, Isabelle De Vincenzib, Philippe Gaillardb,
Cécile Camesa, Paulin Faoc, Stanley Luchtersd,e,f, Nigel Rollinsg,h,
Marie-Louise Newelli, Kirsten Borka, Jennifer S. Readj,k,
for the Kesho Bora Study GroupM
Objective: To assess the relationship between infant feeding practices and mortality by
18 months of age among children born to HIV-infected mothers in the Kesho Bora trial
(Burkina-Faso, Kenya and South Africa).
Methods: Enrolled HIV-infected women were counseled to choose between breastfeeding up to 6 months or replacement feeding from delivery. Multivariable Cox models
were used to compare the infant mortality risks according to feeding practices over time
defined as never breastfed, weaned or still breastfed. The category ‘still breastfed’ was
disaggregated as exclusively, predominantly or partially breastfed to compare modes of
breastfeeding. The relationship between weaning and mortality was also assessed using
marginal structural models to control for time-dependent confounders, such as maternal
or infant morbidity (reverse causality).
Results: Among 795 mothers, 618 (77.7%) initiated breastfeeding. Mortality rates by 18
months among uninfected and infected children were 6 and 38%, respectively. Never
breastfed and weaned children were at greater risk of death compared with those still
breastfed. Adjusted hazard ratios were 6.7 [95% confidence interval (CI)¼2.5–17.9;
P < 0.001] and 6.9 (CI ¼ 2.8–17.2; P < 0.001) for never breastfed and weaned children,
respectively. Estimation of the effect of weaning using marginal structural models led to
similar results. No statistically significant differences were observed according to mode
of breastfeeding (exclusive, predominant or partial).
Conclusion: Within 6 months after birth, weaned or never breastfed children were at
about seven-fold higher risk of dying compared with children who were still breastfed
despite a context in which interventions were provided to reduce risks associated with
ß 2013 Wolters Kluwer Health | Lippincott Williams & Wilkins
replacement feeding.
AIDS 2013, 27:1621–1630
Keywords: Africa, breastfeeding, HIV-exposed children, marginal structural
models, mortality, replacement feeding
a
Institut de Recherche pour le Développement, University of Montpellier, Montpellier, France, bWHO, Reproductive Health and
Research, Geneva, Switzerland, cCentre Muraz, Bobo-Dioulasso, Burkina Faso, dInternational Centre for Reproductive Health
(ICRH), Mombasa, Kenya, eInternational Centre for Reproductive Health, Ghent University, Ghent, Belgium, fCentre for
International Health, Burnet Institute, Melbourne, Victoria, Australia, gUniversity of KwaZulu-Natal, Durban, South Africa,
h
Department of Maternal, Newborn, Child and Adolescent Health, World Health Organization, Geneva, Switzerland, iAfrica
Centre for Health and Population Studies, University of KwaZulu-Natal, Somkhele, South Africa, jEunice Kennedy Shriver
National Institute of Child Health and Human Development, National Institutes of Health, Bethesda, Maryland, and kNVPO/
OASH/OS/DHHS, Washington, DC, USA.
Correspondence to Amandine Cournil, Institut de Recherche pour le Développement, BP 64501, 34394 Montpellier Cedex 5,
France.
Tel: +33 4 67 41 63 22; fax: +33 4 67 41 63 30; e-mail: [email protected]
The Kesho Bora Study Group members are listed in the Appendix.
Received: 21 August 2012; revised: 26 November 2012; accepted: 29 November 2012.
DOI:10.1097/QAD.0b013e32835d5226
ISSN 0269-9370 Q 2013 Wolters Kluwer Health | Lippincott Williams & Wilkins
1621
Copyright © Lippincott Williams & Wilkins. Unauthorized reproduction of this article is prohibited.
1622
AIDS
2013, Vol 27 No 10
Introduction
Methods
Without interventions, the risk of breastmilk transmission of HIV after 4–6 weeks of age is about 0.8% per
month of breastfeeding [1,2]. Complete avoidance of
breastfeeding is the only way of totally preventing
mother-to-child transmission (MTCT) of HIV through
breastmilk. However, avoidance of breastfeeding and
early cessation have been associated with increased risks
of morbidity and mortality in resource-constrained
settings [3–11]. Recently, clinical trials have demonstrated the efficacy of antiretroviral prophylaxis in
reducing the risk of MTCT through breastmilk
[12–16]. Providing maternal or infant antiretroviral
drugs while breastfeeding thus appears to be a promising
strategy to keep children alive while remaining HIVuninfected in resource-constrained settings where
formula feeding is not safe [17].
The design of the Kesho Bora RCT has been described in
detail previously [12,18]. Briefly, pregnant women were
enrolled at antenatal clinics at five study sites (BoboDioulasso, Burkina Faso; Mombasa, Kenya; Nairobi,
Kenya; Durban, South Africa and Somkhele, South
Africa). Enrolment criteria for the RCT were gestational
age below 34 weeks, WHO clinical stage 1, 2, or 3, and a
CD4 cell count of 200–500 cells/ml. Women were
randomized to initiate either triple antiretroviral prophylaxis [combination of zidovudine (ZDV), lamiduvine,
and lopinavir/ritonavir; ¼ triple ARV] until cessation of
breastfeeding or the South-African standard antiretroviral
prophylaxis for MTCT [ZDV until delivery with single
dose nevirapine (NVP) to women at the onset of labour
and to the newborn infant; ¼ ZDV/sdNVP].
Meanwhile, several aspects of the association between
infant feeding modality and mortality need further
investigation. First, the contribution of infant and
maternal morbidity as a determinant of both breastfeeding cessation and death of the child (i.e. reverse
causality) has not been fully addressed. Second, it is
unclear to what extent mortality is a consequence of a
deleterious effect of the event of weaning or of the fact
of receiving replacement feeding per se. Third, little
information is available regarding the association
between the degree of exclusivity of breastfeeding
and mortality.
The present analysis aimed to address these issues using
data collected in the Kesho Bora randomized controlled
trial (RCT), conducted in three countries of subSaharan Africa. This trial assessed the efficacy of
antiretroviral prophylaxis to reduce MTCT of HIV
and followed mother-infant pairs from birth to
12–24 months postpartum with collection of detailed
data on infant feeding practices and morbidity. Women
were counselled to either breastfeed exclusively up to
6 months or to formula-feed from birth, but
nonexclusive breastfeeding was common and many
women breastfed for considerably less than 6 months.
Feeding practices within the first 6 months were highly
heterogeneous, providing a good opportunity to assess
the relationship between mortality and infant feeding
mode.
Specific objectives were: to compare mortality risks
according to breastfeeding practices defined as never
breastfed, weaned (stopped all breastfeeding) or still
breastfed and according to breastfeeding type namely,
exclusive, predominant and partial breastfeeding; and to
assess whether child’s or mother’s health events prior
to breastfeeding cessation contributed to any observed
associations between breastfeeding cessation and
mortality.
Ethical clearance for the study was given by the ethical
and regulatory committees in Burkina Faso, Kenya, and
South Africa, and at the WHO and the US Centers for
Disease Control and Prevention. All women provided
written consent. The RCT was registered with Current
Controlled Trials, ISRCTN71468401.
Infant feeding counselling based on 2004 WHO
guidelines [19] was provided individually to all women
by dedicated nutritionists or nurse counsellors [20,21].
Mother–infant pairs were seen at birth, every 2 weeks
until 8 weeks after delivery, monthly until 12 months and
every 3 months until 18 months to assess clinical,
nutritional and biological characteristics. Infant-feeding
patterns were assessed by interviewers, who were not
involved in infant feeding counselling (with the exception
of Bobo-Dioulasso study site) using an adaptation of the
WHO infant-feeding assessment tool [22]. At each visit,
mothers were asked if their child has been given
breastmilk, replacement feeding or both. From the
2-week to the 6-month visit, an additional questionnaire
was used for breastfeeding women to record foods or
fluids ever given since the last scheduled visit.
Inclusion criteria and definitions for this analysis
Women enrolled in the RCT and their live born infants
were included in the analysis if at least one visit for which
infant feeding was reported and at least one HIV PCR test
result for the child was available. The mode of feeding was
first defined as never breastfed for children who received
only replacement feeding from birth and as ever breastfed
for children who were ever breastfed, irrespective of the
duration of breastfeeding. Second, ever-breastfed children were classified at each point of time as still breastfed
or weaned. The time at weaning was defined as the last
visit at which breastfeeding was reported. Children who
died or were lost to follow-up before weaning were
considered to have been breastfed up to the date of death
(for those who died) or the date of the last contact (for
Copyright © Lippincott Williams & Wilkins. Unauthorized reproduction of this article is prohibited.
Mortality and infant feeding practices Cournil et al.
those lost to follow-up). Third, still breastfed children
were further classified at each timepoint from birth to
6 months as exclusively, predominantly or partially
breastfed according to which fluids and foods they
received [23–25]. A child who was breastfed at a given
age while having never received any other food or drink
with the exception of drops or syrups consisting of
vitamins, mineral supplements or medicines, was
considered as exclusively breastfed. A child who was
breastfed at a given age having ever received food-based
fluid, solid-food or nonhuman milk was considered as
partially breastfed. Breastfed children having been given
water or water-based drinks only were predominantly
breastfed [26,27].
A protected source of water was defined as available if
piped water into the house or compound was the main
source of drinking water. A socioeconomic score was
calculated by multiple correspondence analysis using
eight household assets [28,29]. This score was categorized
in tertiles of increasing economic level within each study
site. Three periods of enrollment were defined: 2005–06,
2007 and 2008. Serious adverse events (SAEs) were
defined as previously described [12]. Low birth weight
was defined as a birth weight below 2500 g. Cotrimoxazole prophylaxis was defined as until 12 months versus
less than 12 months (including none).
Statistical analysis
Maternal and infant characteristics were compared
between ‘never’ and ‘ever’ breastfed groups using
Student’s t-test and the x2-test. Kaplan–Meier survival
analyses were used to estimate cumulative child mortality.
Log rank test was used to compare mortality between
infant feeding groups (never versus ever breastfed).
Cox proportional hazards survival analyses were used to
assess association of infant mortality with infant feeding
patterns. Hazard ratios (HRs) and adjusted hazard ratios
(aHRs) and 95% confidence intervals (CIs) were
obtained. Infant-feeding mode was included as a fixed
variable with two modalities (ever and never breastfed)
[model 1]. Then, feeding practice (still breastfed, weaned
and never breastfed) was tested as a time-varying variable
[model 2]. In a third analysis, the modality ‘still breastfed’
was further split into three categories: exclusively,
predominantly or partially ‘still breastfed’ [model 3].
Model three was run for the first 6 months of age in
which these different feeding practices are defined. In
addition to infant feeding variable, any available baseline
variable (as listed in Table 1) was included in the models if
its association with mortality in univariate analysis had a
P-value below 0.25. Study sites were included in all
models as a fixed effect. Missing data were considered as a
specific category in all analyses to ensure stability of the
sample sizes. Piece-wise time-varying covariates for age at
complete breastfeeding cessation were introduced to
assess whether the association between weaning and
1623
mortality were constant over time. Three age groups were
considered: 0–2.9; 3–4.9; and 5 months and above.
Interactions between any ‘baseline’ variable and infantfeeding variables were added in the model to identify
effect modifiers. Some analyses were done separately
according to HIV-infection status of the child, treatment
allocation (triple ARV vs. ZDV/sdNVP prophylaxis) and
cotrimoxazole prophylaxis, even in the absence of
significant interaction terms. Estimated survival curves
according to feeding practices (time-varying variables)
were obtained using the Simon and Makuch method [30]
(Fig. 1).
To address reverse causality, we reviewed data (i.e.
routinely collected maternal recall of any infant feeding
modification due to the illness leading to death) from the
SAE form reporting the death of the child to determine
whether the mother reported that she changed the way
she fed her child as a result of health events. We also
applied marginal structural models to estimate the effect
of breastfeeding cessation on children’s deaths by
appropriate control for the effects of time-dependent
confounders (i.e. the child’s or the mother’s health events)
[31,32]. Three time-dependent covariates corresponding
to the occurrence of an SAE for the child (excluding
deaths) or for the mother and occurrence of a health event
(diarrhea, vomiting or fever) for the child were
considered. First, logistic regression models were fitted
to predict patient-specific and time-specific probabilities
of weaning as a function of baseline variables (as listed in
Table 1) and health event variables as defined above.
These predicted probabilities were then used to derive
inverse probability weights (IPWs) for weaning. Second,
these IPWs were used in a weighted pooled logistic
model, which included baseline confounders and
breastfeeding cessation variable, to estimate the
parameters of a proportional hazards marginal structural
model. The assumption of positivity was assessed by
checking that all categories of exposure (weaned or not)
were observed for the most important confounders. We
also checked that stabilized weights satisfied the condition
of small variability.
SAS (version 9.2; Cary, North Carolina, USA) and Stata
(version 10.1; Stata Corp, College Station, Texas, USA)
statistical software were used to conduct these analyses.
Results
Size and characteristics of the study population
Of 824 HIV-infected women enrolled in the Kesho Bora
RCT, 805 delivered live born infants between June 2005
and August 2008. Of these 805 infants, six died before
feeding initiation, two were lost to follow-up before
recording information on the first feeding, and two died
without a PCR test to determine HIV-infection status.
Copyright © Lippincott Williams & Wilkins. Unauthorized reproduction of this article is prohibited.
1624
AIDS
2013, Vol 27 No 10
Table 1. Characteristics of the study population of 795 mother–child pairs enrolled in the Kesho Bora randomized controlled trial, according to
mode of feeding.
Never breastfed n ¼ 177
Study site
Bobo Dioulasso, Burkina Faso
Durban, South Africa
Mombasa, Kenya
Nairobi, Kenya
Somkhele, South Africa
Period of enrollment
2005–2006
2007
2008
Maternal characteristics at enrollment
Maternal age (years), median (IQR)
<25
25–34
35
Education level
None
Primary
Secondary
Protected source of water
Yes
No
Socioeconomic score
Low
Medium
High
Treatment allocation
Triple ARV
ZDV/sdNVP
Maternal characteristics at delivery
WHO clinical stage at delivery
1
2
3
CD4 count (cells/ml)
CD4 count median (IQR)
350
<350
Unknownb
Viral load (log10 copies per ml)
Viral load, median (IQR)
Viral load (copies per ml)
300
<300
Unknownb
Mode of delivery
Vaginal
Elective caesarean
Emergency caesarean
Infants characteristics
Sex
Male
Female
Birth weight (g)
2500
<2500
Unknownb
Cotrimoxazole until 12 months
Yesc
No
Infant events by 18 months
HIV infection
Death
Lost-to-follow-up
Ever breastfed n ¼ 618
Pa
16
79
53
10
19
6.6%
43.2%
22.7%
23.3%
20.0%
225
104
180
33
76
93.4%
56.8%
77.3%
76.7%
80.0%
<0.001
59
76
42
19.3%
23.5%
25.4%
247
248
123
80.7%
76.5%
74.6%
0.25
177
54
105
18
27 (24;32)
21.4%
22.6%
23.1%
618
198
360
60
27 (23;31)
78.6%
77.4%
76.9%
0.24
0.26
6
45
126
5.0%
16.5%
31.3%
113
228
277
95.0%
83.5%
68.7%
<0.001
113
64
27.0%
17.0%
305
313
73.0%
83.0%
<0.001
51
60
66
18.9%
22.9%
25.1%
219
202
197
81.1%
77.1%
74.9%
0.22
90
87
22.8%
21.7%
304
314
77.2%
78.3%
0.70
128
41
8
23.1%
21.1%
17.0%
427
152
39
76.9%
78.8%
83.0%
0.58
171
112
59
6
415 (325;515)
20.2%
31.6%
11.3%
571
443
128
47
454 (359;584)
79.8%
68.4%
88.7%
0.03
0.001
170
2.5 (2.2;3.1)
560
2.7 (2.2;3.5)
0.03
86
84
7
22.0%
24.8%
10.8%
305
255
58
78.0%
75.2%
89.2%
0.37
141
9
27
20.0%
30.0%
44.3%
563
21
34
80.0%
70.0%
55.7%
<0.001
92
85
23.8%
20.8%
294
324
76.2%
79.2%
0.30
162
12
3
23.6%
17.9%
7.1%
524
55
39
76.4%
82.1%
92.9%
0.29
72
105
40.7%
59.3%
516
102
83.5
16.5
10
18
19
5.6%
10.2%
10.7%
52
45
80
8.4%
7.3%
12.9%
<0.001
0.23
0.21
0.43
Note: Data are shown as number and percentage unless otherwise indicated.
P-value for comparison between never and ever-breastfed infants.
The category ‘unknown’ was not considered in the comparison of the distribution between infant-feeding modes.
c
Yes, if the child received cotrimoxazole prophylaxis until 12 months or until the month preceding death or lost to follow-up.
a
b
Copyright © Lippincott Williams & Wilkins. Unauthorized reproduction of this article is prohibited.
(95% CI)
Model 3b,c
0.96
(0.2;13.2)
(0.3;36.5)
(3.3;71.2)
(2.5;59.4)
Still breastfed
0.98
Weaned
Mortality and infant feeding practices over time
The median duration of breastfeeding was 4.9 months
(interquartile range, 2.8;6.0). A total of 161 (26.1%)
<0.001
<0.001
HR
(2.8;17.2)
(2.5;17.9)
P
(95% CI)
0.20
<0.001
<0.001
0.43
0.34
<0.001
0.004
(0.8;2.5)
(2.2;13.3)
(1.8;11.1)
(0.3;15.7)
(0.3;35.7)
(3.1;63.5)
(2.0;41.5)
1
1.6
(0.9;3.0)
0.13
1
6.9
6.7
HR
1
1.4
1
5.4
4.5
1
2.2
3.2
14.0
9.2
In univariate or multivariate analyses, study site, maternal
age, maternal education, source of water, socioeconomic
score, treatment allocation, maternal HIV-related characteristics at delivery (CD4 cell count and viral load), mode
of delivery, infant characteristics (sex, birth weight) and
cotrimoxazole prophylaxis were not associated with
mortality. Period of enrollment showed a weak association with mortality in univariate analysis (see table,
supplemental digital content 1, http://links.lww.com/
QAD/A294). The risk of death tended to be higher for
never-breastfed children compared with ever-breastfed
children (Table 2, Model 1). HIV infection status of the
child was strongly associated with the probability of death
(aHR, 8.0; CI, 4.5;14.2; P < 0.001).
Ever breastfed
Never breastfed
Still breastfed
Weaned
Never breastfed
Exclusively breastfedc
Predominantly breastfed
Partially breastfed
Weaned
Never breastfed
Mortality among ever and never breastfed infants
A total of 63 children died within 18 months. The
estimated overall mortality rate was 4.5% by 6 months,
7.4% by 12 months, and 8.3% by 18 months of age. The
estimated cumulative mortality rate by 18 months was
10.5% in the never breastfed group compared with 7.7%
among ever breastfed children, but the difference was not
statistically significant (P-value for log-rank test ¼ 0.20).
P
Therefore, the study population for this analysis
comprised 795 mother-infant pairs. Of these, 177
(22.3%) mothers never breastfed, whereas 618 (77.7%)
initiated breastfeeding (Table 1). Compared with mothers
who initiated breastfeeding, mothers who chose replacement feeding had a higher level of education and were
more likely to have access to a protected source of water.
They were also more likely to have a CD4þcell countless
than 350 cells/ml and to have had a Caesarean delivery.
Cumulative follow-up rates at the 18-month visit
(excluding deaths) were similar in formula fed and
ever-breastfed children (89.3 versus 87.1%; P ¼ 0.43).
(95% CI)
Fig. 1. Estimated survival curves according to feeding
practices using the Simon and Makuch method [30].
HR
Time (months)
P
12
(95% CI)
10
HR
8
Model 2b
6
Model 1b
4
Multivariatea
2
Univariate
0
Table 2. Association between infant feeding practices and mortality among 795 children born to HIV-infected mothers enrolled in the Kesho Bora trial.
Never breastfed
0.92
1
1.8
3.2
15.2
12.2
0.94
1625
All analyses include 795 children. Missing data for CD4 cell count, viral load and birth weight were considered as a specific category.
a
Models 1 and 2 are adjusted for study site, period of enrollment, source of water socioeconomic score, treatment allocation, CD4 cell count, viral load, mode of delivery and child’s HIV infection;
Model 3 is adjusted for study site, period of enrollment, socioeconomic score, treatment allocation, CD4 cell count and child’s HIV infection.
b
Model 1 includes a fixed variable for mode of infant feeding; models 2 and 3 include a time-varying variables with three categories in model 2 (still breastfed, weaned and never breastfed) and five
categories in model 3 (exclusively, predominantly, partially breastfed, weaned and never breastfed).
c
For detailed feeding practices in univariate analyses or in model 3, analyses were run for mortality in the first 6 months.
P
1.0
0.55
0.34
<0.001
0.002
Mortality and infant feeding practices Cournil et al.
Copyright © Lippincott Williams & Wilkins. Unauthorized reproduction of this article is prohibited.
1626
AIDS
2013, Vol 27 No 10
mothers stopped breastfeeding before their child reached
3 months of age, 162 (26.2%) mothers continued to
breastfeed at 6 months of age. Weaned and neverbreastfed children were at greater risk of death compared
with still-breastfed children with similar adjusted HRs
(Table 2, Model 2 and Fig. 1).
introducing two time-varying covariates for these two
periods in comparison with weaning from 5 months of age.
Among ever breastfed children, weaning before 3 months
was associated with a 3.5-fold increase in mortality (CI:
1.7;7.3; P < 0.001), whereas weaning at 3 to 4.9 months
was associated with a 2.2-fold increase (CI: 0.9–5.4;
P ¼ 0.08) compared with weaning after 5 months.
Interaction terms combining study arm, HIV infection
status of the child or cotrimoxazole use with infant feeding
modality were not statistically significant (P > 0.20) and
HR estimates in adjusted analyses were similar across the
different subgroups. The aHR of weaned versus still
breastfed was 8.7 (CI: 2.7–28.5; P < 0.001) and aHR of
never breastfed versus still breastfed was 7.8 (CI: 2.3–26.7;
P ¼ 0.001) among uninfected children. In infected
children aHR of weaned versus still breastfed was equal
to 4.8 (CI: 1.1–20.3; P ¼ 0.03) and aHR of never breastfed
versus still breastfed was 5.5 (CI: 0.9–31.6; P ¼ 0.06). The
excess of mortality among weaned or never-breastfed
children compared with still breastfed was also observed in
the ZDV/sdNVP arm [aHR (weaned): 8.9; CI: 2.7;29.8;
P < 0.001; aHR(never breastfed): 8.7; CI: 2.4;31.1;
P < 0.001] and to a lesser extent in the triple ARV arm
[aHR (weaned): 4.4; CI: 1.1;17.8; P ¼ 0.04; aHR (never
breastfed): 3.7; CI: 0.8;16.4; P ¼ 0.09]. Results were less
homogeneous when stratifying by cotrimoxazole prophylaxis. An excess of mortality among weaned or never
breastfed children compared with still breastfed children
was found in those who received cotrimoxazole prophylaxis until 12 months [aHR (weaned): 6.2; CI: 2.2;17.3;
P < 0.001; aHR(never breastfed): 5.7; CI: 1.7;18.9;
P < 0.01] but not among children who did not receive
the prophylaxis throughout infancy [aHR (weaned): 0.8;
CI: 0.1;8.5; P ¼ 0.83; aHR (never breastfed): 1.3; CI:
0.2;9.3; P ¼ 0.76].
Reverse causality and time-varying confounders
Breastfeeding cessation as a consequence of a serious
illness that led to death was reported in one case. Overall,
the median period of time between breastfeeding
cessation and death was 2.4 months [interquartile range
(IQR): 1.4–5.4] and only one child died within 15 days
after breastfeeding cessation.
Marginal structural models were used to account for timedependent confounding. For ever-breastfed children, the
unweighted model (i.e. without controlling for health
event confounding) indicated that weaned children had a
higher risk of mortality compared with still breastfed with
an adjusted HR of 5.0 (CI: 1.8–13.6). The aHR obtained
from the full weighted-model was slightly attenuated [4.4
(CI: 1.6–11.9)]. Similar HR attenuation was observed in
uninfected and infected children (data not shown).
Mortality and mode of breastfeeding over time
Of 618 women, who initiated breastfeeding, 246 (39.8%)
and 137 (22.2%) were still exclusively breastfeeding at 3
and 5 months, respectively. One hundred and four
(16.8%) practiced partial breastfeeding for more than 1
month, in the first 6 months. When disaggregated by
breastfeeding type, there was no statistical difference in
the risk of death within the first 6 months between
children who were exclusively, predominantly or partially
breastfed (Table 2, Model 3). Weaned and never-breastfed
children were at a substantially higher risk of mortality in
comparison with exclusively breastfed children.
Gastroenteritis and pneumonia accounted for 71% of all
causes of death in weaned and formula fed children,
whereas gastroenteritis was never reported as a cause of
death in children who were still breastfed at the time of
death (Table 3).
Discussion
To investigate whether the association between weaning
and mortality varied with age, we assessed the effect of
weaning before 3 months and between 3 and 4.9 months by
The relationship between infant feeding practices
and mortality was investigated in 795 children born to
Table 3. Causes of death according to HIV-infection status and mode of feeding at the time of death.
HIV-uninfected children
HIV-infected children
Weaned/never breastfed
Still breastfed
Weaned/never breastfed
Still breastfed
12
15
1
6
–
3
37
0
2
0
2
–
0
4
10
2
2
0
3
1
18
0
1
1
0
1
0
3
Gastroenteritis
Pneumonia
Other infections
Malaria
AIDS/HIV
Others
Alla
a
Cause was missing for one death.
Copyright © Lippincott Williams & Wilkins. Unauthorized reproduction of this article is prohibited.
Mortality and infant feeding practices Cournil et al.
HIV-infected mothers in three countries of sub-Saharan
Africa. No evidence of difference in mortality risks was
found when comparing never breastfed to ever-breastfed
children. If mode of feeding was taken into account using
time-dependent variables, weaned or never-breastfed
children were at about a seven-fold higher risk of dying
compared with children who were still breastfed.
Additional results indicated that the adverse effect
associated with breastfeeding cessation decreased with
increasing age of infant between birth and 6 months of
age. Moreover, although evaluating associations in subsamples like infection status or trial arm may have been
limited by the small number of events, breastfeeding
appears to remain protective for HIV-infected infants as
for HIV-uninfected ones, and whether or not the mother
receives antiretroviral prophylaxis during breastfeeding.
Previous studies documenting an association between
feeding modality and mortality among HIV-exposed
children have shown that, at best, replacement feeding
results in no benefit in survival [7,33–37] and, at worse,
results in a six-fold increase in mortality by 12 months of
age compared with breastfeeding [3,5]. Heterogeneity of
findings between these reported studies can be explained
by different socioeconomic and cultural settings. The
definition of infant-feeding modality (fixed or timedependent variable) used in studies might also have led to
the different magnitude of effect; however the direction
of effect was consistent in the various studies.
In agreement with observations made by Kuhn and
colleagues [11], our results indicated a continuous adverse
effect of receiving replacement feeding after cessation of
breastfeeding, rather than a temporary deleterious effect
of the event of weaning. First, deaths in weaned children
were not concentrated in the period immediately
following weaning but were homogeneously distributed
in the whole period of replacement and complementary
feeding. Second, HRs were similar in weaned and never
breastfed children. The median time period between
breastfeeding cessation and death was 2.4 months, which
is longer than in a study in rural Uganda (i.e. 1.5 months)
despite similar duration of follow-up, maybe because the
level of schooling and hygiene were lower in the latter [3].
No evidence of difference between exclusive and
predominant breastfeeding was found, and partial
breastfeeding was associated with an intermediate risk
between exclusive/predominant and replacement feeding. However, the number of events was limited in the
partial category and the study was not powered to
examine this question. Several studies have documented
the relationship between breastfeeding mode and HIV
transmission or HIV-free survival in HIV-exposed
children [38–42] whereas others focused specifically on
morbidity or mortality of children [3,9,39,43]. Overall,
these studies indicated that exclusive breastfeeding is
associated with the lowest risks of transmission, morbidity
1627
and mortality, and partial breastfeeding with the highest,
but the difference between these two categories was
reduced for mortality risk.
An important strength of this study lies in the effort made
to control for confounding, notably of time-dependent
variables. Indeed, the study of infant-feeding modality
and mortality may be affected by a time-dependent
confounding, which cannot be handled by traditional
statistical means. Few studies have addressed this issue of
reverse causality [11,44,45]. Some investigators tried to
account for situations in which breastfeeding cessation
and death both caused by a serious illness occurred within
a short time, about 1 week. In this analysis, we addressed
the issue of reverse causality by using marginal structural
models. It allowed us to account for potential reverse
causality situations wherein an illness episode, breastfeeding cessation, and death could occur over a long period.
Although the child’s and the mother’s morbidity were
determinants of weaning (results not shown), the amount
of confounding was low. Marginal structural models are
appropriate tools to explore causal relationships. However, estimation of the causal effect relies on several
assumptions including the assumption of exchangeability,
that is, all confounders are measured and appropriately
taken into account in the models. Although we cannot
exclude the possibility of unmeasured confounding, the
great amount of information collected during the Kesho
Bora trial allowed us to include all major potential
confounders in the analysis.
It should be pointed out that breastfeeding durations
were limited to approximately 6 months. Women were
strongly encouraged to wean by that age in agreement
with international and national recommendations at the
time of the study. Our results cannot necessarily be
generalized to longer periods of breastfeeding. However,
it seems reasonable to hypothesize that the decrease in the
protective effect of breastfeeding with increasing child
age demonstrated within the 0–6-month age interval
continues beyond the age of 6 months.
Four out of five of the Kesho Bora study sites were in
urban African settings. Our results cannot be generalized
to rural areas. However, it is likely that the detrimental
effect of non or short-duration breastfeeding would be
even larger in more remote rural areas.
In conclusion, the data presented here confirm the
benefits of breastfeeding for survival of HIV-exposed
children for the first 6 months in a context in which
extensive counseling, support and other interventions
were provided to reduce risks. Additionally, they confirm
the potential risks of replacement feeding, independent of
the event of weaning. In resource-constrained settings, it
has long been recognized that promotion of breastfeeding
is a key intervention for reducing the mortality among
HIV-exposed and nonexposed children. In the context of
Copyright © Lippincott Williams & Wilkins. Unauthorized reproduction of this article is prohibited.
1628
AIDS
2013, Vol 27 No 10
antiretroviral drugs being available to reduce postnatal
transmission through breastmilk, interventions to support
HIV-infected mothers to exclusively breastfeed should be
prioritized to promote HIV-free survival of HIV-exposed
children.
Acknowledgements
We thank all participants enrolled in the study. We
acknowledge Pierre de Beaudrap for statistical advice.
Author contribution: J.S.R., A.C., I.D.V., K.B. conceived
the analysis. A.C. performed the analysis with the
contribution of P.G., C.C. and K.B. for infant feeding
definitions implementation. A.C., J.S.R., K.B., I.D.V.
wrote the article. All authors read, commented and
approved the final version of the article.
The Kesho Bora Study Group and the study sites: Bobo
Dioulasso, Burkina Faso (Centre Muraz): Nicolas Meda
(Principal Investigator), Paulin Fao, Odette Ky-Zerbo,
Clarisse Gouem (Study coordinators), Paulin Somda,
Hervé Hien, Patrice Elysée Ouedraogo, Dramane Kania,
Armande Sanou, Ida Ayassou Kossiwavi, Bintou Sanogo,
Moussa Ouedraogo, Issa Siribie (Investigators), Diane
Valéa (Laboratory Coordinator), Sayouba Ouedraogo &
Roseline Somé (Data Managers), François Rouet (InterSite Laboratory Coordination); Durban, South Africa
(University of KwaZulu Natal): Nigel Rollins (Principal
Investigator), Lynne McFetridge, Kevi Naidu (Study
Coordinators); Mombasa, Kenya (International Centre
for Reproductive Health): Stanley Luchters, Marcel
Reyners (Principal Investigators), Eunice Irungu (Study
Coordinator), Christine Katingima, Mary Mwaura and
Gina Ouattara (Investigators), Kishor Mandaliya, Sammy
Wambua (Laboratory Coordinators), Mary Thiongo
(Data Manager); Nairobi, Kenya (Network for AIDS
Researchers in East and Southern Africa): Ruth Nduati
(Principal Investigator), Judith Kose (Study Coordinator),
Ephantus Njagi (Laboratory Coordinator), Peter Mwaura
(Data Manager). Somkhele, South Africa (Africa Centre
for Health and Population Studies, University of
KwaZulu Natal): Marie-Louise Newell (Principal Investigator), Stephen Mepham (Study Coordinator),
Johannes Viljoen (Laboratory Coordinator), Ruth Bland
(Investigator), Londiwe Mthethwa (Data Manager).
Supporting institutions: Agence Nationale de
Recherches sur le SIDA et les hépatites virales, France:
Brigitte Bazin and Claire Rekacewicz (Sponsor Representatives); Centers for Disease Control and Prevention,
USA: Allan Taylor (Sponsor Representative and CoInvestigator), Nicole Flowers, Michael Thigpen, Mary
Glenn Fowler, Denise Jamieson (Co-Investigators);
Eunice Kennedy Shriver National Institute of Child
Health and Human Development, National Institutes
of Health, USA: Lynne M. Mofenson (Sponsor
Representative), Jennifer S. Read (Co-Investigator);
Institut de Recherche pour le Développement (IRD),
Montpellier, France: Kirsten Bork, Cécile Cames and
Amandine Cournil (Nutrition Coordination); International Centre for Reproductive Health (ICRH), Ghent
University, Ghent, Belgium: Patricia Claeys, Marleen
Temmerman, Stanley Luchters (Sponsor Representatives); Université Montpellier 1, EA 4205 ‘Transmission,
Pathogenèse et Prévention de l’infection par le VIH’ ; and
CHU Montpellier, Laboratoire de Bactériologie-Virologie, Montpellier, France: Philippe Van de Perre, Pierre
Becquart (until December 2006), Vincent Foulongne,
Michel Segondy (Laboratory Coordination).
Study coordination: World Health Organization, Geneva, Switzerland: Isabelle de Vincenzi (Study Coordinator), Philippe Gaillard (Site Coordinator), Tim Farley
(Project Manager), Ndema Habib (Study Statistician),
Sihem Landoulsi (Study Analyst).
Financial support was provided by Agence Nationale de
Recherches sur le SIDA et les hépatites virales (ANRS),
Department for International Development (DFID),
European and Developing Countries Clinical Trials
Partnership (EDCTP), Thrasher Research Fund, Belgian
Directorate General for International Cooperation,
GlaxoSmithKline Foundation, Centers for Disease
Control and Prevention, Eunice Kennedy Shriver
National Institute of Child Health and Human Development, UNDP/UNFPA/UNICEF/World Bank/
WHO Special Programme of Research, Development
and Research Training in Human Reproduction and the
Victorian Operational Infrastructure Support Program.
Funding: The Bobo-Dioulasso site was funded by
l’Agence Nationale de Recherches sur le Sida et les
Hépatites Virales (ANRS) and UNDP/UNFPA/World
Bank/WHO Special Programme of Research, Development and Research Training in Human Reproduction
(WHO/HRP).
The Mombasa site was funded by ANRS, WHO/HRP,
European and Developing Countries Clinical Trials
Partnership (EDCTP), Thrasher Research Fund, Belgian
Directorate General for International Cooperation.
The Nairobi site was funded by the Centers for Disease
Control and Prevention (CDC) and the Eunice Kennedy
Shriver National Institute of Child Health and Human
Development (NICHD) through a cooperative agreement.
The South-African sites were funded by the Department
for International Development (DFID), EDCTP,
UNICEF and WHO/HRP.
The Nutrition and laboratory coordination were funded
by ANRS.
Copyright © Lippincott Williams & Wilkins. Unauthorized reproduction of this article is prohibited.
Mortality and infant feeding practices Cournil et al.
The overall coordination and external monitoring was
funded by WHO/HRP.
The findings and conclusions in this report are those of
the authors and do not necessarily represent the views of
the World Health Organization, of the Centers for
Disease Control and Prevention or of the National
Institutes of Health
Conflicts of interest
The authors have no conflict of interest to disclose.
References
1. The Breastfeeding and HIV International Transmission Study
Group. Late postnatal transmission of HIV-1 in breast-fed
children: an individual patient data meta-analysis. J Infect
Dis 2004, 189:2154–166.
2. Kourtis AP, Bulterys M. Human immunodeficiency virus type 1
(HIV-1) and breastfeeding. New York: Springer US; 2012.
3. Homsy J, Moore D, Barasa A, Were W, Likicho C, Waiswa B,
et al. Breastfeeding, mother-to-child HIV transmission, and
mortality among infants born to HIV-Infected women on highly
active antiretroviral therapy in rural Uganda. J Acquir Immune
Defic Syndr 2010; 53:28–35.
4. Kafulafula G, Hoover DR, Taha TE, Thigpen M, Li Q, Fowler
MG, et al. Frequency of gastroenteritis and gastroenteritisassociated mortality with early weaning in HIV-1-uninfected
children born to HIV-infected women in Malawi. J Acquir
Immune Defic Syndr 2010; 53:6–13.
5. Kagaayi J, Gray RH, Brahmbhatt H, Kigozi G, Nalugoda F,
Wabwire-Mangen F, et al. Survival of infants born to HIVpositive mothers, by feeding modality, in Rakai, Uganda. PLoS
One 2008; 3:e3877.
6. Kuhn L, Aldrovandi G. Survival and health benefits of breastfeeding versus artificial feeding in infants of HIV-infected
women: developing versus developed world. Clin Perinatol
2010; 37:843–862.
7. Kuhn L, Aldrovandi GM, Sinkala M, Kankasa C, Semrau K,
Mwiya M, et al. Effects of early, abrupt weaning on HIV-free
survival of children in Zambia. N Engl J Med 2008; 359:130–
141.
8. Taha TE, Hoover DR, Chen S, Kumwenda NI, Mipando L,
Nkanaunena K, et al. Effects of cessation of breastfeeding in
HIV-1-exposed, uninfected children in Malawi. Clin Infect Dis
2011; 53:388–395.
9. Taha TE, Kumwenda NI, Hoover DR, Kafulafula G, Fiscus SA,
Nkhoma C, et al. The impact of breastfeeding on the health of
HIV-positive mothers and their children in sub-Saharan Africa.
Bull World Health Organ 2006; 84:546–554.
10. Thior I, Lockman S, Smeaton LM, Shapiro RL, Wester C,
Heymann SJ, et al. Breastfeeding plus infant zidovudine prophylaxis for 6 months vs formula feeding plus infant zidovudine
for 1 month to reduce mother-to-child HIV transmission in
Botswana: a randomized trial: the Mashi Study. JAMA 2006;
296:794–805.
11. Kuhn L, Sinkala M, Semrau K, Kankasa C, Kasonde P, Mwiya M,
et al. Elevations in mortality associated with weaning persist
into the second year of life among uninfected children born to
HIV-infected mothers. Clin Infect Dis 2010; 50:437–444.
12. Kesho Bora Study Group. Triple antiretroviral compared with
zidovudine and single-dose nevirapine prophylaxis during
pregnancy and breastfeeding for prevention of mother-to-child
transmission of HIV-1 (Kesho Bora study): a randomised controlled trial. Lancet Infect Dis 2011; 11:171–180.
13. Tonwe-Gold B, Ekouevi DK, Viho I, Amani-Bosse C, Toure S,
Coffie PA, et al. Antiretroviral treatment and prevention
of peripartum and postnatal HIV transmission in West
Africa: evaluation of a two-tiered approach. PLoS Med 2007;
4:e257.
1629
14. Chasela CS, Hudgens MG, Jamieson DJ, Kayira D, Hosseinipour
MC, Kourtis AP, et al. Maternal or infant antiretroviral drugs to
reduce HIV-1 transmission. N Engl J Med 2010; 362:2271–
2281.
15. Kumwenda NI, Hoover DR, Mofenson LM, Thigpen MC,
Kafulafula G, Li Q, et al. Extended antiretroviral prophylaxis
to reduce breast-milk HIV-1 transmission. N Engl J Med 2008;
359:119–129.
16. Shapiro RL, Hughes MD, Ogwu A, Kitch D, Lockman S, Moffat
C, et al. Antiretroviral regimens in pregnancy and breastfeeding in Botswana. N Engl J Med 2010; 362:2282–2294.
17. Horvath T, Madi BC, Iuppa IM, Kennedy GE, Rutherford GW,
Read JS. Interventions for preventing late postnatal mother-tochild transmission of HIV. Cochrane Database Syst Rev
2009:CD006734. DOI: 10.1002/14651858.CD006734.pub2
18. Kesho Bora Study Group. Safety and effectiveness of antiretroviral drugs during pregnancy, delivery and breastfeeding for
prevention of mother-to-child transmission of HIV-1: the
Kesho Bora Multicentre Collaborative Study rationale, design,
and implementation challenges. Contemp Clin Trials 2010;
32:74–85.
19. WHO/UNICEF/UNAIDS/UNFPA. HIV and infant feeding: a
guide for healthcare managers and supervisors. 2004. http://
whqlibdoc.who.int/hq/2003/9241591234.pdf, [Accessed 18
July 2012].
20. Bork K, Cames C, Cournil A, Musyoka F, Ayassou KA, Naidu K,
et al. Infant feeding modes and determinants among HIV-1infected women in the Kesho-Bora study. J Acquir Immune
Defic Syndr 2013; 62:109–118.
21. Cames C, Saher A, Ayassou KA, Cournil A, Meda N, Simondon
KB. Acceptability and feasibility of infant-feeding options:
experiences of HIV-infected mothers in the World Health
Organization Kesho Bora mother-to-child transmission prevention (PMTCT) trial in Burkina Faso. Matern Child Nutr
2010; 6:253–265.
22. WHO. Breastfeeding and replacement feeding pratices in the
context of mother-to-child transmission of HIV. An assessment
tool for research and programs. 2001. http://whqlibdoc.who.int/
hq/2001/WHO_CAH_01.21.pdf, [Accessed 18 July 2012].
23. WHO. Indicators for assessing breastfeeding practices. 1991.
http://whqlibdoc.who.int/hq/1991/WHO_CDD_SER_91.14.pdf,
[Accessed 18 July 2012].
24. WHO WHO/UNICEF. Breastfeeding counselling: a training
course. 1993. (http://www.who.int/maternal_child_adolescent/
documents/who_cdr_93_3/en/index.html [Accessed 18 July
2012]
25. WHO. Indicators for assessing infant and young child feeding
practices – Part 1: Definitions. 2008. (http://whqlibdoc.who.int/
publications/2008/9789241596664_eng.pdf [Accessed 18 July
2012]
26. Becquet R, Ekouevi DK, Menan H, Amani-Bosse C, Bequet L,
Viho I, et al. Early mixed feeding and breastfeeding beyond 6
months increase the risk of postnatal HIV transmission: ANRS
1201/1202 Ditrame Plus, Abidjan, Cote d’Ivoire. Prev Med
2008; 47:27–33.
27. WHO. HIV transmission through breastfeeding: a review of
available evidence. update 2007. (http://whqlibdoc.who.int/
publications/2008/9789241596596_eng.pdf [Accessed 18 July
2012]
28. Mejean C, Traissac P, Eymard-Duvernay S, El Ati J, Delpeuch F,
Maire B. Influence of socio-economic and lifestyle factors
on overweight and nutrition-related diseases among Tunisian
migrants versus nonmigrant Tunisians and French. BMC Public
Health 2007; 7:265.
29. Delpeuch F, Cornu A, Massamba JP, Traissac P, Maire B. Is body
mass index sensitively related to socio-economic status and to
economic adjustment? A case study from the Congo. Eur J Clin
Nutr 1994; 48 (Suppl 3):S141–S147.
30. Schultz LR, Peterson EL, Breslau N. Graphing survival curve
estimates for time-dependent covariates. Int J Methods
Psychiatr Res 2002; 11:68–74.
31. Fewell Z, Hernan MA, Wolfe F, Tilling K, Choi H, Sterne JAC.
Controlling for time-dependent confounding using marginal
structural models. Stata J 2004; 4:402–420.
32. Hernan MA, Brumback B, Robins JM. Marginal structural
models to estimate the causal effect of zidovudine on the
survival of HIV-positive men. Epidemiology 2000; 11:561–
570.
Copyright © Lippincott Williams & Wilkins. Unauthorized reproduction of this article is prohibited.
1630
AIDS
2013, Vol 27 No 10
33. Becquet R, Bequet L, Ekouevi DK, Viho I, Sakarovitch C,
Fassinou P, et al. Two-year morbidity-mortality and alternatives to prolonged breast-feeding among children born to HIVinfected mothers in Cote d’Ivoire. PLoS Med 2007; 4:e17.
34. Mbori-Ngacha D, Nduati R, John G, Reilly M, Richardson B,
Mwatha A, et al. Morbidity and mortality in breastfed and
formula-fed infants of HIV-1-infected women: a randomized
clinical trial. JAMA 2001; 286:2413–2420.
35. Newell ML, Coovadia H, Cortina-Borja M, Rollins N, Gaillard P,
Dabis F. Mortality of infected and uninfected infants born to
HIV-infected mothers in Africa: a pooled analysis. Lancet 2004;
364:1236–1243.
36. Peltier CA, Ndayisaba GF, Lepage P, van Griensven J, Leroy V,
Pharm CO, et al. Breastfeeding with maternal antiretroviral
therapy or formula feeding to prevent HIV postnatal motherto-child transmission in Rwanda. AIDS 2009; 23:2415–2423.
37. Rollins NC, Becquet R, Bland RM, Coutsoudis A, Coovadia HM,
Newell ML. Infant feeding, HIV transmission and mortality at
18 months: the need for appropriate choices by mothers and
prioritization within programmes. AIDS 2008; 22:2349–2357.
38. Becquet R, Bland R, Leroy V, Rollins NC, Ekouevi DK, Coutsoudis A, et al. Duration, pattern of breastfeeding and postnatal
transmission of HIV: pooled analysis of individual data from
West and South African cohorts. PLoS One 2009; 4:e7397.
39. Coovadia HM, Rollins NC, Bland RM, Little K, Coutsoudis A,
Bennish ML, Newell ML. Mother-to-child transmission of HIV-1
infection during exclusive breastfeeding in the first 6 months of
life: an intervention cohort study. Lancet 2007; 369:1107–1116.
40. Coutsoudis A, Pillay K, Kuhn L, Spooner E, Tsai WY,
Coovadia HM. Method of feeding and transmission of HIV-1
from mothers to children by 15 months of age: prospective
cohort study from Durban, South Africa. AIDS 2001; 15:379–
387.
41. Coutsoudis A, Pillay K, Spooner E, Kuhn L, Coovadia HM.
Influence of infant-feeding patterns on early mother-to-child
transmission of HIV-1 in Durban, South Africa: a prospective
cohort study. South African Vitamin A Study Group. Lancet
1999; 354:471–476.
42. Iliff PJ, Piwoz EG, Tavengwa NV, Zunguza CD, Marinda ET,
Nathoo KJ, et al. Early exclusive breastfeeding reduces the risk
of postnatal HIV-1 transmission and increases HIV-free survival. AIDS 2005; 19:699–708.
43. Fawzy A, Arpadi S, Kankasa C, Sinkala M, Mwiya M, Thea DM,
et al. Early weaning increases diarrhea morbidity and mortality
among uninfected children born to HIV-infected mothers in
Zambia. J Infect Dis 2011; 203:1222–1230.
44. 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; 355:451–
455.
45. Bahl R, Frost C, Kirkwood BR, Edmond K, Martines J,
Bhandari N, Arthur P. Infant feeding patterns and risks of
death and hospitalization in the first half of infancy: multicentre cohort study. Bull World Health Organ 2005; 83:418–
426.
Copyright © Lippincott Williams & Wilkins. Unauthorized reproduction of this article is prohibited.