Nduati’s claim that breastfeeding HIV-positive mothers have higher mortality may be unwarranted

To the Lancet Editors:

Nduati et al claim to show that breastfeeding dramatically increases mortality among HIV+ mothers [1]. The commentary by Newell [2] discusses some of the limitations of this study, and compares it to research by Coutsoudis et al [3] that draws very different conclusions. We would like to address additional problems with the paper.

Randomization to achieve baseline equivalence in sample groups is an essential element of a clinical trial. There are several reasons to believe that randomization was not successful in this research. A more detailed description of the trial [4] notes that, at birth, only 2 of the babies in the formula fed arm were believed to be HIV-positive, compared to 9 assigned to the breastfeeding arm. Obviously, method of feeding cannot have any impact on HIV status at the time of birth, and these results may indicate important differences between the mothers in the treatment arms that were not detected by the researchers nor balanced by the randomization.

Comparisons of breastfeeding with other feeding methods obviously cannot be blinded. Consequently, the study team would know which feeding method each mother and baby had been allocated. This introduces a potential source of bias, as the attitudes of researchers can influence the participants in the trial. In addition, it appears that the formula feeding mothers may have had more extensive contact with researchers, in order to educate them about the proper way to prepare formula [4]. This may have led, if only indirectly, to more health care interventions for these women.

Compliance and definition of breastfeeding are other serious issues. Unlike Coutsoudis, Nduati’s research does not distinguish between partial and exclusive breastfeeding, and does not precisely define exclusive breastfeeding. It does document that 9% of mothers were exclusively breastfeeding by 6 months. This is much higher than the 3.5% at 4-5 months in the general population found in the Demographic & Health Survey in 1998 in Kenya [5], perhaps partly because the WHO definition of exclusive breastfeeding was not used in the Nduati study. This would exclude even giving water.

Compliance with exclusive formula feeding was only 71% [1], meaning that 29% of the mothers in this arm practiced partial breastfeeding like those in the ‘breastfeeding’ arm. Given that compliance was based on self-reporting the figure of 71% is probably an over-estimate.

Because the analysis was intent-to-treat, it is not clear how different the two arms of the trial were in actual feeding practices, as both contained a significant number of mixed feeders. It is quite possible that some mothers in the ‘formula feeding’ arm of the trial breastfed more than some mothers in the ‘breastfeeding arm’.

Nduati et al claim that they have no reason to believe that dropouts from the trial (which were greater in number than the deaths) differed between arms of the trial, but it can equally be said that there is no reason to believe that they are the same. It is conceivable, for example, that a seriously ill woman would find the burden of preparing formula for a baby too great and be more likely to drop out. Conversely, it is possible that some of the mothers who were healthiest and of highest socioeconomic status in the breastfeeding arm may have believed that their babies were at a disadvantage and left the trial to give them the perceived advantage of formula feeding. It is not possible to draw firm conclusions from a trial in which the outcomes (maternal death) are smaller than the anomalies in the trial (e.g. loss to follow-up, no vital status information available).

We are concerned that despite potentially serious methodological flaws in the work by Nduati et al, policy recommendations may be based on its conclusions, ignoring research by Coutsoudis et al that concludes that there is no risk to the health of mother (or child) from exclusive breastfeeding.

An additional concern is the lack of written, witnessed consent for participation. Surely, even if some of these women could not read and needed verbal information, they could still all sign their own name. We request further clarification in the pages of Lancet or on their website, providing justification for the use of oral consent.

Policy decisions based on this information may seriously impact the lives of millions of mothers and children. We request that the research team led by Nduati make their raw data available to other researchers for re-analysis.

For AnotherLook, a not-for-profit organization dedicated to gathering information, raising critical questions and stimulating needed research about infant feeding in the context of HIV/AIDS.

Marian Tompson
Executive Director AnotherLook

Authors

Phyll Buchanan, UK; David Crowe, Canada; Ted Greiner, Sweden; George Kent, USA; Tessa Martyn, UK; Valerie McClain, USA; Pamela Morrison, Zimbabwe; Denise Parker, USA; Magda Sachs, UK; Marian Tompson, USA; Karen Zeretzke, USA

References

[1] Nduati R et al. Effect of breastfeeding on mortality among HIV-1 infected women: a randomised trial. Lancet. 2001 May 26; 357: 1651-5.

[2] Newell ML. Does breastfeeding really affect mortality among HIV-1 infected women? Lancet. 2001 May 26; 357: 1634-5.

[3] Coutsoudis A et al. Are HIV-infected women who breastfeed at increased risk of mortality? AIDS. 2001 Mar 30; 15(5): 653-5.

[4] Nduati R et al. Effect of breastfeeding and formula feeding on transmission of HIV-1; a randomized clinical trial. JAMA. 2000 Mar 1; 283: 1167-74.

[5] Kenya Central Statistical Bureau and Macro International, Inc., 1998. Demographic and Health Survey. Calverton MD: Macro International, Inc.