One of the principal gains in maternal and child health during the past few decades has been the revival of breastfeeding. The immediate and long-term benefits of breastfeeding have been well documented [1–4]. The HIV epidemic, however, has made it impossible to have a uniform policy on breastfeeding for rich and poor countries. There is clear evidence to show that HIV is transmitted through breast milk [5–7] and the increased risk of vertical transmission by this route is estimated to be about 14% for women with established infection . Accordingly, in 1992, the World Health Organization (WHO) made recommendations on breastfeeding based on whether infectious diseases and malnutrition were or were not the primary causes of infant deaths among populations ; in developing countries HIV-infected women were encouraged to continue breastfeeding. In a more recent statement, UNAIDS has given far greater emphasis to ensuring individual informed choice  in accordance with the recognition that there are circumstances in developing countries where alternatives to breastfeeding for HIV-seropositive women are a safe option . Such circumstances might apply to countries with intermediate economies, such as South Africa. Implications of breastfeeding under conditions of varying infant mortality rates, vertical transmission rates through breastfeeding, and HIV prevalence and incidence have been modelled, and these calculations may be useful to countries developing their own policies . Although the dangers of transmission of virus have been recognized, there is little information on whether breast milk of HIV-seropositive women is as protective against common childhood infections, malnutrition, progression to AIDS, and early death, as that of HIV-seronegative women. A recent editorial concluded that the lack of data on whether breast- or bottle-feeding by HIV-seropositive mothers affected overall outcome in their babies, limited womens' choices in deciding on feeding method .
We report on the association between feeding practice and outcome in a cohort of children born to HIV-1-infected women from Durban, South Africa.
Materials and methods
This report is derived from a prospective, hospital based, cohort study on the natural history of vertically transmitted HIV-1 infection. The study was conducted at King Edward VIII hospital, a large urban hospital in Durban, KwaZulu/Natal, South Africa. About 99% of the patient population is black. The major causes of infant death in KwaZulu/Natal are infections, malnutrition, and perinatal problems [14,15]. South Africa is classified as a middle income country by the World Bank with a per capita income of US$2670.
The sampling method has previously been described in detail . All women attending the hospital's antenatal clinic for the first time in the index pregnancy, were given counselling on HIV infection and tested if informed consent was obtained. The acceptance rate was over 95%. The prevalence of HIV-1 infection in the hospital's antenatal clinic was 10% during the study period. Post-test counselling was given once results were obtained. Liveborn infants of infected women were entered into the study with maternal permission. Every mother who was approached agreed to allow her newborn to participate in the study.
All infants were examined within 48 h of birth and measurements taken of length, weight, and head circumference. The infants were followed up at monthly intervals for the first 3 months, at 3 month intervals up to 24 months, and thereafter at 6 month intervals. At each visit, a detailed clinical examination was performed and growth measurements taken. A record was made of any interim illnesses and visits to health centres, and of hospital admissions. Mothers were encouraged to bring the child to the follow-up clinic for any complaint, so that episodes of illness would not be missed.
At no stage of the study were the mothers influenced against breastfeeding. The policy in the neonatal unit during the study period was to advise mothers on the advantages of breastfeeding. Women who chose to feed their infants on formula received advice on hygienic preparation of feeds and were not provided with any financial support for artificial feeding. Method of feeding was recorded in detail for the first 12 months; thereafter this assessment became difficult as the infants were either fed totally on family diet or complementary feeds made up the major part of the diet. Feeding method was defined as: exclusive breastfeeding, where the child was on breast-feeds only from birth (these infants received no supplementary milk feeds); mixed feeding, where the child was receiving both formula feeds and breast milk (the period of exclusive breastfeeding and the age at which formula was commenced were noted, as well as the duration during which the infant received both breast and formula feeds); and exclusive formula feeding, where the infant received formula feeds only. All groups of infants would have received other complementary foods for varying periods.
The infants received no antiretroviral or immunotherapy during the course of the study.
Classification of infection status was made according to the recommendations of the Ghent workshop . Infants were regarded as infected if they were antibody positive at 15 months or had an HIV-related death. They were classified as non-infected if the antibody test was negative from 9 months of age, or if death was non-HIV-related. Those infants who were lost to follow-up before the age of 9 months whilst still antibody-positive and those whose cause of death could not be determined, were classified as indeterminate. The diagnosis of AIDS was based on the revised WHO criteria .
Definition of morbidity
Diarrhoea, pneumonia, and otitis media were chosen as outcomes because these are the infections that breastfeeding appears to prevent in non-HIV-infected populations, and because diarrhoea and pneumonia are the common causes of morbidity and mortality in black South African infants and children. The following criteria were used to define morbidity: diarrhoea, three or more episodes of loose stool per day, present for at least 3 days; pneumonia, the presence of tachypnoea and crackles and/or radiological changes; failure to thrive, weight and length below the third percentile or crossing of percentile lines ; otitis media, inflammation of the eardrum or a purulent discharge from the ear.
Blood samples were taken at regular intervals from birth. These were tested for HIV-1 antibodies by a commercial enzyme-linked immunosorbent assay (ELISA; Abbot Laboratories, North Chicago, Illinois, USA), by a confirmatory Roche ELISA (Cobus Core, Basel, Switzerland) or an immunoflourescent assay (IFA; Virion, Cham, Switzerland). Samples were considered positive if the second ELISA or the IFA was positive. The same method was applied for maternal samples.
Social and demographic characteristics were compared between the three feeding method groups using analysis of variance or χ2 test. The association between feeding and rate of HIV infection was assessed for maternal risk factors which could impact on transmission, including age, parity, haemoglobin level, and mode of delivery [20–22]. The latter two factors were included because these were found to be significantly associated with transmission in a previous report on this population . Logistic regression analysis was used to control for the effect of these variables on the transmission rate and feeding method.
For mortality and progression to AIDS, survival function estimates were computed using the product-limit (Kaplan–Meier) method. The curves were compared using the log-rank test.
Morbidity was expressed, according to feeding practice and HIV status, as incidence densities per 100 child months of follow-up.
Between October 1990 and April 1993, 234 black infants and their 229 mothers were entered into the study. Nine women refused entry into the study. Fifty-three patients did not attend a single follow-up visit and were excluded from the study. This group was not different from the rest of the cohort with regard to maternal characteristics used to compare the groups.
The remaining 181 infants were classified as 48 infected (including 17 deaths); 93 non-infected, and 40 indeterminate (including eight deaths). The median vertical transmission rate was 34% [confidence interval (CI), 26–42%]. The upper estimate, which assumes all indeterminates were infected, was 48% (40–60%) and the lower estimate, which assumes all indeterminates were non-infected, was 26.5% (20–33%).
The mean follow-up period for the breastfed infants was 22.8 months, for those on mixed feeding it was 22.5 months, and for those on formula only it was 20.6 months.
HIV-infected and non-infected infants
In the first part of the analysis, we have excluded the indeterminate group.
Feeding and transmission
There were 133 infants for whom adequate feeding information was available. Twenty-one infants (16%) were fed exclusively on formula, 36 infants (27%) were exclusively breastfed and 76 (57%) received both breast and formula (mixed) feeds. The median duration of exclusive breastfeeding for the whole group was 5.0 months (range, 1–12 months), for those exclusively breastfed it was 12.0 months (range, 1–12 months), and for those on mixed feeds it was 2.0 months (range, 1–12 months).
When the relationship between feeding and transmission was analysed more closely (Table 1), it was found that infants who were exclusively formula-fed had a lower transmission rate (24%), than those infants who received either mixed feeding (32%) or exclusive breastfeeding (39%); the relative risk for infection in the exclusively breastfed versus those on formula only, was 1.63 (CI, 0.7–−3.76; P = 0.24). The increased risk for transmission by breastfeeding compared to formula feeding was 15% (CI, 1.8–31.8).
Maternal risk factors which could impact on transmission were compared and were found not to be different between the three groups. Mode of delivery, maternal age, and maternal haemoglobin level had no impact on transmission rate by feeding group. When these were controlled for in a logistic regression model, the association between increased risk and exclusive breastfeeding remained, but was of borderline significance [relative risk (RR), 10.7; CI, 1.1–96.0].
To determine whether duration of breastfeeding affected transmission, we assessed the risk of infection according to the duration of exclusive breastfeeding (to the nearest month). The infection rates were 45% of those infants receiving breast milk only for the first month, 64% of those receiving breast milk only for 2 months of life, and 75% of those receiving breast milk only for 3 months. However there were only four infants in the latter group. Thereafter the rate remained constant at ± 25% for up to 12 months of exclusive breastfeeding. These figures are not statistically significant.
Feeding and mortality
Deaths occurred only in the HIV-infected infants. Of the 17 infected infants who died, seven were exclusively breastfed and 10 had mixed feeding. No deaths occurred in the exclusively formula-fed group during the study period, compared to a mortality of seven out of 36 (19%) in the exclusively breastfed infants, and of 10 out of 76 (13%) in the infants receiving mixed feeding (RR, 1.87, CI, 0.61–6.66). The RR of death in the exclusively breastfed group compared to that in the exclusively formula-fed group, could not be defined, as there were no deaths in those exclusively formula-fed.
When we compared feeding versus mortality in the HIV-infected infants only, we found mortality to be highest in the exclusively breastfed infants seven out of 14 (50%), compared to 10 out of 24 (42%) in the infants receiving mixed feeding and 0 out of 5 (0%) in those infants receiving formula only. However, survival at 12 months was similar in the exclusively breastfed infants to those on mixed feeding (64% versus 65%).
Feeding and morbidity
Table 2 shows the relationship between feeding method and morbidity. The data shown represent a total of 2924 child-months of follow-up. We found no significant difference between the group that was exclusively breastfed and those who received mixed feeding and exclusive formula feeding. This was similar for the HIV-infected and the non-infected infants.
Feeding and AIDS
Among the infected infants, seven out of 14 (50%) of those exclusively breastfed, 13 out of 24 (54.1%) on mixed feeding, and none out of four (0%) on formula only, developed AIDS during the study period. Infants who were exclusively breastfed appeared to progress to AIDS more slowly than those who received mixed feeds: mean age at diagnosis of AIDS was 7.5 months in the breastfed group (SD, 5.3; range, 2–18 months) compared to 5.0 months (SD, 2.0; range, 3–10 months) in the mixed feeding group (P = 0.2242).
The indeterminate group of infants
A similar pattern for morbidity to that detected in the infected and non-infected groups as described above, was noted; little information was available for the seven deaths in this group.
In this prospective study of a cohort of babies born to black HIV-1-infected South African women, we did not find the expected benefits of exclusive breastfeeding. We employed four outcome criteria to measure benefit: mortality, infections, failure to thrive, and progression to AIDS. Of the HIV-infected infants who died during the study period, the mortality was highest in those exclusively breastfed, although this did not reach statistical significance. Our study does not allow us to determine the cause of this finding. This ambiguity may be the result of research design: the study did not set out to test the effects of breastfeeding against formula by a randomized controlled method. We have calculated that the sample size in our population (ante-natal HIV-1 prevalence 20%) to detect differences in mortality between breast- and formula-fed infants of HIV-seropositive mothers will be 2500 antenatal attendees (80% power, 5% significance level). Disturbances in breast milk immune factors directed at common infections of infancy and the impact of HIV virions on gastrointestinal integrity require further investigation.
Morbidity due to common childhood infections (diarrhoea, pneumonia, otitis media) and failure to thrive, were as frequent in babies who were exclusively breastfed as in those who were either given formula exclusively or a mix of formula and breast milk; this observation was true for both HIV-infected and non-infected infants and children.
To the best of our knowledge there are four other sets of similar data, some of which are comparable to the results given here. Results from an Italian study  showed that breastfeeding was of short-term advantage: progression to AIDS was slower and survival longer in breastfed as opposed to formula-fed HIV-infected infants. This advantage of breastfeeding was lost when children were 5 years of age . However, data were retrospectively collected and the duration of breastfeeding was not known. In a study of relatively privileged women from the Republic of Congo (formerly Zaire), breastfeeding was found to protect the infants from common childhood illnesses, in those born to both HIV-infected and HIV-negative women , but the HIV-infected infants were not compared by feeding method and all those who died within the first 6 months were excluded from the analysis. The closest findings to those presented here are also from Africa: in a prospective study from Nairobi, breastfeeding by HIV-infected women for longer than 15 months, was more often associated with growth retardation than breastfeeding for shorter periods . Gray et al. reported an increased transmission rate and an increased mortality in exclusively breastfed HIV-infected infants from Soweto, South Africa . In breastfed infants, mortality occurred in nine out of 114 infants (7.0%), compared to one out of 49 (2%), in infants on mixed feeding (P = 0.14). Moreover, there were no adverse effects on growth or morbidity, and no increase in hospital admissions among breastfed over non-breastfed infants born to HIV-seropositive women.
The higher transmission rate in the exclusively breastfed group compared to either the mixed breast- and formula-fed group or to those exclusively formula-fed, is similar to findings summarized in a recent meta-analysis on this subject . There are several studies from the developed countries as well as from Africa that have reported on the association between breastfeeding and transmission of HIV [24,25,20]. An increased risk of transmission through breastfeeding has been reported from Brazil, a country with an intermediate economy . The Italian study also showed that the risk of transmission was increased with breastfeeding and that the risk increased with the duration of breastfeeding . The data from Durban do not reach statistical significance but are suggestive of a similar direction in the relationship between HIV transmission and duration of breastfeeding.
The greater prevalence of infections and failure to thrive among the HIV-infected infants was predictable and extends the results from other studies in developing  and industrialized countries . It is essential that the trend towards an association between breastfeeding and HIV-related mortality (detected in both Durban and Johannesburg) and the absence of data for HIV-unrelated deaths (given the small number of deaths in the indeterminate group) be explored further.
Recent reviews of the available published data on the subject of breastfeeding have confirmed once again that breastfeeding protects against infections and reduces infant mortality; the conclusion was drawn that the relative risk for mortality of artificial feeding was between one and nine . These findings generally apply to developing and developed countries. However, most studies on breastfeeding were undertaken in the pre-HIV era. As the benefits of breastfeeding were well established, we did not include a control group of HIV-seronegative pregnant women and their offspring.
This study has several shortcomings which require explanation. The women were not randomly allocated to breastfeeding versus non-breastfeeding groups; they self-selected their feeding method. It has been argued, among key research scientists, that randomized studies in poor countries will be unethical  given the undisputed public health benefits of breastfeeding and the prospects of allocating women without access to adequate sanitation and clean water to a non-breastfeeding group, and women with such access to breastfeeding with its concomitant risk of HIV transmission. Secondly, the number of women who chose to feed by formula exclusively was small; therefore, the number of HIV-infected infants in the artificially fed group was minuscule. This, in fact, is the real position in Africa; very few women feed by formula only. The pattern of feeding adopted by the women in this study is similar to that for most women in Africa . In addition, we were unable to assess the possible role of other contributing factors to the results obtained including maternal viral load, CD4/CD8 counts, and p24 antigen levels . We could not assess whether the differences in time to AIDS between breastfed and formula-fed infants were due to dissimilarities in the proportions infected in utero, intrapartum, and postnatally. We have data which suggest that the proportions of infection in our cohort were 27% and 72% during delivery and postnatally (unpublished data). The delay to AIDS in the breastfed infants may have been due to these being infected by breast milk in contrast to those on formula who could have been infected earlier. It should be noted that the comparison of time to AIDS according to feeding method was limited, as we did not control for confounding factors.
The inclusion of the indeterminate group of infants is unlikely to have altered the main conclusions of the study. The proportion of these infants in the different feeding groups resembled that in the 141 infants classified as infected or non-infected. The maternal characteristics, newborn and early infancy features in this indeterminate group were similar to those included for analysis. Moreover, morbidity patterns were similar, and the available information suggests that there were no differences in morbidity between the breastfed infants and those receiving formula or mixed feeds; the seven deaths in this group may have added clarity to the mortality differences between the groups.
In a recent workshop on breastfeeding in HIV-positive women (held in Durban, South Africa), health professionals from central, east, and southern Africa agreed that there was no conflict of purpose in safeguarding individual choice while promoting breastfeeding at the population level. The results from this study do not contradict this consensus view, and they support the recent UNAIDS recommendations .
In conclusion, our findings have particular relevance to populations in countries with intermediate economies where the under-5 years mortality rates straddle the division between high and moderate , and HIV-infected women can choose safe alternatives to breastfeeding. In these situations, all pregnant women should be offered counselling and HIV testing, and those found to be HIV-positive should be advised that breastfeeding may not provide the anticipated degree of protection against common infections and growth failure, and is associated with a higher risk of transmission. We expect that breastfeeding will regain its role as the pre-eminent method of infant feeding, in individual countries, as the epidemic declines.
The authors thank E. Gouws and C. Connolly, Department of Biostatistics, Medical Research Council of South Africa, for the statistical analysis and Z. Stein and L. Kuhn, Sergievsky Center, Columbia University, New York, USA for helpful comments, especially on statistical analysis, on the first draft of this manuscript.
1. Howie PW, Forsyth JS, Ogston SA, Clark A, Florey CD: Protective effect of breast feeding against infection
2. Plank SJ, Milanese ML: Infant feeding and infant mortality in Chile
. Bull WHO
3. Victoria CG, Smith PG, Vaughan JP, et al.
: Evidence for protection by breast feeding against infant deaths from infectious diseases in Brazil
4. Cunningham AS: Morbidity in breast fed and artificially fed infants
. J Pediatr
5. Douglas GC, King BF: Maternal-fetal transmission of human immunodeficiency virus: a review of possible routes and cellular mechanisms of infection
. Clin Infect Dis
6. Van de Perre P, Simonon A, Hitimana DG, et al.
: Infective and anti-infective properties of breastmilk from HIV-1 infected women
7. Van de Perre P: Postnatal transmission of human immunodeficiency virus type 1: the breastfeeding dilemma
. Am J Obstet Gynecol
8. Dunn DT, Newell ML, Ades AE, Peckham CS: Risk of human immunodeficiency virus type 1 transmission through breastfeeding
9. World Health Organization Global Programme on AIDS: Consensus statement from the WHO/UNICEF consultation on HIV transmission and breastfeeding
. Weekly Epidemiol Rec
10. UNAIDS: HIV and infant feeding: an interim statement
. Wkly Epidemiol Rec
11. Nicoll A, Newell M-L, Van Praag E, Van de Perre P, Peckham C: Infant feeding policy and practice in the presence of HIV-1 infection
12. Hu DJ, Heyward WL, Byers Jr RH, et al.
: HIV infection and breastfeeding: policy implications through a decision analysis model
13. Ziegler JB: Breast feeding and HIV
14. Bradsaw D, Buthelezi G: Health status
. In South African Health Review
. Edited by Harrison D, Barron P, Edwards J, Durban: Health Systems Trust and Henry J Kaiser Foundation; 1996, 19–32.
15. South African Vitamin A Consultative Group: Anthropometric, vitamin A, iron and immunisation coverage status in children aged 6–71 months in South Africa,1994
. S Afr Med J
16. Bobat RA, Coutsoudis A, Coovadia HM, Moodley D: Determinants of mother to child transmission of HIV1 infection in a cohort from Durban, South Africa
. Pediatr Infect Dis J
17. Dabis F, Msellati P, Dunn D, et al.
: Estimating the rate of mother-to-child transmission of HIV. Report of a workshop on methodological issue, Ghent (Belgium), 1992
18. World Health Organization Global Programme on AIDS: Report of the Meeting of the Technical Working Group on HIV/AIDS in Childhood
. Geneva: WHO; 1989.
19. National Centre for Health Statistics. Monthly Vital Statistics Report
. Columbia: Ross Laboratories; 1976, 25
(3) (suppl HRA): 76–120.
20. European Collaborative Study: Risk factors for mother to child transmission of HIV-1
21. Borkowsky W, Krasinski K: Perinatal human immunodeficiency virus infection: ruminations on mechanisms of transmission and methods of interventions
22. Gabiano C, Tovo PA, de Martino M, et al.
: Mother-to-child transmission of human immunodeficiency virus type 1: risk of infection and correlates of transmission
23. Tozzi AE, Pezzotti P, Greco D: Does breastfeeding delay progression to AIDS in HIV infected children? [letter] AIDS
24. De Martino M, Tovo P-A, Tozzi AE, et al.
: HIV-1 transmission through breast-milk: appraisal of risk according to duration of feeding
25. Ryder RW, Manzila T, Baende E, et al.
: Evidence from Zaire that breastfeeding by HIV-1-seropositive mothers is not a major route for perinatal HIV-1 transmission but does decrease morbidity
26. Datta P, Embree JE, Kreiss JK, et al.
: Mother to child transmission of human immunodeficiency virus type 1: report from the Nairobi study
. J Infect Dis
27. Gray GE, McIntyre JA, Lyons SF: The effect of breastfeeding on vertical transmission of HIV-1 in Soweto, South Africa
. XIth International Conference on AIDS
. Vancouver, July 1996 [abstract ThC.415].
28. Collareda NL, Frota AC, Silva-Oliviera RH, Schechter M: Maternal clinical status, breastfeeding, and vertical transmission of HIV-1 in Rio de Janeiro
. Pediatr AIDS HIV Infect Fetus Adolesc
29. De Martino M: The Italian register for HIV infection in children. Human immunodeficiency virus type 1 infection and breast milk
. Acta Paediatr
1994 (suppl 400): S51-S58.
30. Mgone CS, Mhalu FS, Shao JF, et al.
: Prevalence of HIV-1 infection and symptomatology of AIDS in severely malnourished children in Dar ES Salaam, Tanzania
. J Acquir Immune Defic Syndr
31. European Collaborative Study: Children born to women with HIV-1 infection: natural history and risk of infection
32. Peckham C, Gibb D: Mother-to-child transmission of the human immunodeficiency virus
. N Engl J Med
33. Nakajima H: Statistics
. In The World Health Report, 1996. Fighting Disease, Fostering Development. Report of the Director General
. Geneva: World Health Organization; 1996, annex 3: 199–122.