Does Breastfeeding Affect the Health of HIV-Positive Women? Studies Disagree

Lisa Remez, Guttmacher Institute

First published online:

Two recent studies yield contradictory findings on whether the decision to breastfeed affects the health of HIV-infected women. In one study, a secondary analysis of data from 397 seropositive new mothers in Nairobi, Kenya, women who were randomly assigned to breastfeed were three times as likely to die within 24 months of delivery as were those assigned to formula-feed.1 Moreover, infants born to HIV-infected women who died had an elevated risk of dying before their second birthday, even after the researchers controlled for whether the infant was infected with HIV.

In the other study, also a secondary analysis, data collected from 566 seropositive mothers in Durban, South Africa, showed no significant difference at the univariate level in mortality by 15 months postpartum between women who chose to breastfeed and those who elected to give their infant formula.2 Further, multivariate logistic regression found no significant difference in morbidity, even after baseline CD4 counts and hemoglobin levels were controlled for.

The Kenya Study

The first analysis used data originally collected for a randomized trial that assessed the impact of infant feeding practices on the risk of mother-to-child HIV transmission. Women who attended four prenatal clinics in Nairobi from November 1992 through October 1997 were offered HIV testing; participants who tested positive were randomly assigned to either breastfeed or formula-feed. The infant-mother pairs, who were followed for a maximum of two years, were examined monthly during the first year postpartum and quarterly during the second year. For the mortality analysis, women who failed to keep their follow-up appointments were traced, and the cause of death for those who died was determined from hospital records or from relatives. The researchers used Cox proportional hazards models to calculate the relative risk of maternal death by feeding-group assignment.

At the time of enrollment, there were no significant differences between the two groups by median age (23 years), weight (63 kilograms), number of previous births (one) or HIV-1 RNA viral load (48 virions per mLx103among women who breastfed, and 37 virions per mLx103 among those who formula-fed). However, women assigned to breastfeed were significantly more likely than those assigned to formula-feed to comply with their original feeding instructions (96% vs. 71%).

The final sample for the mortality analysis included 197 mothers randomly assigned to breastfeed (of whom 39 were subsequently lost to follow-up) and 200 women assigned to formula-feed (33 later lost to follow-up). Eighteen of the breastfeeding women died during the two years following delivery (11%), compared with six in the formula-feeding group (4%). Cox proportional hazards models indicate a three-fold relative risk of death for breastfeeding women relative to those using formula (relative risk of 3.2, p=.01). Additional Cox models show that the association between being randomly assigned to breastfeed and dying within two years of delivery remained significant even after the women's viral load and CD4 count at enrollment were controlled for (relative risk of 4.7, p=.001).

Women who were assigned to breastfeed lost more weight in the two-year postpartum period than did those assigned to formula-feed. Weight loss was significantly associated with mortality, as each kilogram that a woman lost per month raised her risk of dying by a factor of 3.4. Adjusting the mortality data for weight loss, however, did not change the overall association between breastfeeding and maternal death.

An HIV-positive mother's CD4 count and her viral load were significantly associated with the risk of death. For example, the risk of dying among women with CD4 counts below 200 cells per mL was 14.7 times as high as the risk among women with counts higher than 500 cells per mL. Further, women whose viral load at enrollment was higher than the median were significantly more likely to die than were women with a viral load below the median (risk ratio, 8.0). The association between being randomly assigned to breastfeed and maternal mortality remained significant even after the researchers controlled for these biological markers of HIV progression.

Finally, an infant's risk of dying, irrespective of the timing of the infant's death relative to the mother's, was significantly higher--by a factor of three--if the mother died; that increase in risk was greater still among babies whose death followed their mother's death (5.6). This association increased in magnitude once the researchers controlled for the infant's HIV-infection status (7.9).

The investigators note that the small absolute number of maternal deaths, the loss-to-follow-up rate of nearly 18% and the fact that these Nairobi women constituted a highly selected subgroup all limit the generalizability of the findings. They hypothesize, moreover, that because nearly one-third of the women in the formula-feeding group did not comply with their assigned feeding practice, the study might underestimate the true risk of maternal death associated with breastfeeding among HIV-positive women.

The investigators offer two biological mechanisms for the possible association between breastfeeding and mortality in HIV-infected mothers. First, the combination of HIV infection and breastfeeding in women who are already inadequately nourished may create an overwhelming metabolic burden. Second, lactation itself may increase the ability of HIV-1 to replicate--either because it raises the level of the hormone prolactin, which may further depress the immune system of an immunocompromised woman, or because factors such as nipple cracking, candida infection or breastmilk production enhance the replication of the virus. The authors conclude that "further research is needed into the mechanism of the association between lactation and maternal death" and into the efficacy of providing HIV-infected women who choose to breastfeed with nutritional supplements to lower their excess risk of mortality.

The author of a related editorial3 points out several problems with interpreting the results of this study, including the fact that breastfeeding's effect on mortality among HIV-infected women was an ad hoc hypothesis; that women randomized to breastfeed had a higher median viral load at enrollment in the study; and that data on women's clinical progress and information on their actual feeding practice--as opposed to the one they were assigned to adopt--are missing. She concludes that the study "would benefit from being complemented by a creative analysis of actual feeding practice to further explore this complex issue."

The South Africa Study

The data used in the second analysis were originally collected in a vitamin A intervention trial among HIV-positive pregnant women. In this study, conducted at two prenatal clinics in Durban, South Africa, 566 women were recruited from July 1995 through April 1998 to randomly receive either a vitamin A supplement or a placebo. The women were asked to make an informed decision on whether to breastfeed or not; 410 women decided to breastfeed and 156 chose to give formula to their baby.

The women were followed up for a mean length of 10.4-10.6 months (range of one month to 18 months). The researchers compared the two groups of women in terms of mortality and seven specific health conditions (pulmonary tuberculosis; candidiasis at more than two sites; pneumonia; ear, nose and throat infections; gastroenteritis; gynecologic infections; and cesarean wound sepsis).

At the univariate level, there was no difference in mortality by 15 months postpartum between women who chose to breastfeed and those who opted to formula-feed (0.5% and 2%, respectively, p=.10). Moreover, there was no difference by feeding choice among all 566 women in the proportions who were diagnosed with any of the seven specific conditions (13% among those who were breastfeeding vs. 15% among those who were not, p>.10). Results of a multivariate logistic regression analysis indicate that the risks of morbidity for breastfeeding and nonbreastfeeding mothers was not significantly different (95% confidence interval, 0.43-1.39), even after baseline CD4 counts and hemoglobin counts were controlled for.

In addition, the researchers collected data at three months postpartum on 10 general clinical symptoms* among 180 women at one clinic (147 who had breastfed their infant and 33 who had not), as well as follow-up CD4 counts and hemoglobin levels for 115 of these women (93 who elected to breastfeed and 22 who chose to formula-feed). Among the 180 women for whom detailed follow-up data on clinical symptoms were available, there was no difference by infant feeding choice in the proportion who reported any symptom (35% vs. 39%, p>.10). Moreover, among the 115 women for whom postpartum CD4 counts and hemoglobin levels were available, there was no significant difference in those counts between women who chose to breastfeed and those who chose to formula-feed (means of 239 and 249, respectively, p>.10); the researchers similarly found no significant difference in three-month postpartum hemoglobin levels by feeding choice (12.0 vs. 11.3, p=.07). According to results of a multivariate linear regression analysis, which controlled for the type of treatment women were assigned to (vitamin A supplement or placebo) and their baseline CD4 counts, there was no significant difference by decision to breastfeed in women's CD4 counts at three months postpartum (p>.10).

The researchers note that their data provide no evidence of "deleterious effects of breastfeeding on the health of seropositive women." They conclude that "the counseling provided to HIV-infected women on feeding choice should continue to be based on current recommendations of UNAIDS."--L. Remez


1. Nduati R et al., Effect of breastfeeding on mortality among HIV-1 infected women: a randomised trial, Lancet, 2001, 357(9269):1651-1655.

2. Coutsoudis A et al., Are HIV-infected women who breastfeed at increased risk of mortality? research letter, AIDS, 2001, 15(5):653-655.

3. Newell M-L, Does breastfeeding really affect mortality among HIV-1 infected women? commentary, Lancet, 2001, 357(9269):1634-1635.