Death of the mother within 18 months after delivery was also significantly associated with lower infant birth weight and infant's vital status and HIV infection (see Table 2B). Of the 162 women who died through 18 months after delivery, 64 had children who also died within this 18-month period. Forty children died before their mothers (for 24 [60%], the time between deaths was <2 months), and 24 children died after their mothers (for 15 [62.5%], the time between deaths was <2 months).
Time to Death Analyses
Proportional hazards regression models were fitted to the data to assess homogeneity of mortality risk, with terms included to represent the main effects of trials grouped by geographic region (East Africa, West Africa, South Africa [reference category]), prophylaxis arm (intervention or placebo), and the interaction of trials with prophylaxis (to allow for the possibility that the effects of prophylaxis differed by geographic region). Neither the prophylaxis arm nor the trial by prophylaxis interaction was significantly associated with mortality, but mortality did differ significantly by geographic region (data not shown). The risks of mortality among women enrolled in trials conducted in East Africa (hazard ratio [HR] = 3.01, 95% CI: 1.86, 4.95) or West Africa (HR = 2.01, 95% CI 1.10, 3.70) were significantly higher than among women in trials in South Africa. Therefore, trials (grouped) were included in the base model used to assess the influence of children's feeding modality and the remaining maternal variables included in Table 2 on HIV-1-infected mothers' mortality.
The influence of each covariate on mothers' mortality was evaluated using the Cox model by fitting a model that included trials (grouped geographically) and the covariate. Only maternal CD4+ count (<200 cells/mm3 vs. ≥200 cells/mm3) was strongly associated with mothers' mortality in the Cox model (P < 0.05). Controlling for differences in mortality by trials (grouped), mothers with lower CD4+ counts were at higher risk of mortality (≤200 cells/mm3: HR = 11.2, 95% CI: 7.7-16.4; >200 cells/mm3: reference category).
With trials (grouped) included in the model, the mortality risk was similar among women who ever or never breast-fed their children (model 1: HR = 0.49, 95% CI: 0.23-1.06; P = 0.07; Table 3). After controlling for the main effect of maternal CD4+ count, the difference remained nonsignificant (model 3: HR = 0.51, 95% CI: 0.20-1.26; P = 0.14). The results reported for model 2 suggest that the relation of children's feeding modality to mothers' mortality essentially did not differ between those with and without maternal CD4+ count available. To assess whether the inclusion of women who only briefly breast-fed their children was influencing the results, the model was run excluding those who breast-fed for <1 month; the results were not appreciably altered. An additional model to assess whether the relation between mothers' mortality and children's feeding modality varied by maternal CD4+ count did not identify a significant interaction (P = 0.59).
To account for the possibility and timing of the mother ceasing breast-feeding during the course of follow-up, proportional hazards analyses were carried out with children's feeding modality treated as a time-dependent covariate (Table 4). Women currently breast-feeding at the time of study events (deaths) were found to be at lower risk of mortality compared with those who had ceased by the time of study events. The addition of maternal CD4+ count to the model resulted in a modest change to the HR for children's feeding modality (model 3: HR = 0.07, 95% CI: 0.04-0.12; P < 0.0001). Models including terms for the interaction of grouped trials with maternal CD4+ count or children's feeding modality (still breast-feeding or not) failed to converge because of a lack of events in some strata. When treated as a time-dependent covariate, a child's vital status, a possible correlate of the mother's HIV disease progression, did not contribute significantly to the model (P = 0.37), nor did it affect the risk estimate associated with children's feeding modality.
Sensitivity analyses were conducted to examine the stability of the final time-dependent covariate model. To assess the influence of possible informative censoring, the final model was fitted under 2 extreme assumptions: that all women lost to follow-up died shortly after they were censored, and that all women lost to follow-up were all alive at 18 months. Estimates under these extreme assumptions supported the primary analysis, although with markedly different HRs for children's feeding modality. Additional models examined children's feeding modality at 1, 3, or 6 months before the mother's death (instead of predicting mortality risk by children's feeding modality at the time of the mother's death). Results of these “lagged” analyses also supported the primary analysis; those mothers still breast-feeding at 1, 3, or 6 months before death continued to have a significantly reduced mortality risk.
In this analysis of more than 4000 HIV-1-infected mothers in sub-Saharan Africa, there were no statistically significant differences in the risk of mortality during the 18-month period after delivery among women who ever breast-fed their children compared with those who never breast-fed their children, with or without adjustment for maternal HIV-1 disease stage (as assessed by CD4+ count) around the time of delivery. Among women who initiated breast-feeding, however, lower risks of mortality were observed with longer durations of breast-feeding. Maternal CD4+ counts were lower among women who never breast-fed or who breast-fed for shorter periods than among those who breast-fed for longer periods. With children's feeding modality treated as a time-dependent covariate, a lower mortality risk was observed among those still breast-feeding compared with those not breast-feeding, with similar results observed with adjustment for maternal CD4+ count. Finally, mothers' mortality was associated with adverse events in the child, including low birth weight, mortality, and HIV-1 infection.
This individual patient data meta-analysis has several strengths, including the large size of the pooled data set (the largest database to evaluate mortality among HIV-1-infected women according to children's feeding modality compiled to date). Selection bias is unlikely, because identification of eligible trials was not based solely on previous publications. Regular and repeated assessments of children's feeding modality and maternal vital status occurred during each of the trials, allowing categorization according to whether or not the mother initiated breast-feeding as well as by duration of breast-feeding and length of time between cessation of breast-feeding and the mother's death. The assessment of the relation between mothers' mortality and children's feeding modality was limited by the available covariates, including those potentially related to choices regarding the children's feeding modality, especially women's reasons for not initiating breast-feeding or for ceasing breast-feeding; the reasons are often complex and inextricably linked to cultural, societal, economic, and family dynamics.14 Also, the amount of breast-feeding (eg, total volume of breast milk ingested over a specified period) was not quantified across all trials. Maternal CD4+ count data were available only around the time of delivery, precluding analysis of maternal CD4+ counts as a time-dependent covariate. Finally, despite the large size of the study population, there were too few deaths to evaluate interactions of covariates formally (eg, to assess whether relations between mortality and feeding modality or CD4+ count varied across geographic regions).
Because the analyses were performed according to children's actual feeding modality, without random assignment to different feeding modalities, characteristics of the mother or of the child could be directly or indirectly associated with the chosen modality. Most importantly, preceding morbidity of the mother, such as complications of delivery, or of the child could affect the mother's decision regarding initiation of breast-feeding or cessation once breast-feeding had been initiated.15 For example, the association between a longer duration of breast-feeding and a lower risk of mothers' mortality theoretically could represent a “protective” effect of breast-feeding on the risk of mortality or, more likely, simply that women who live longer have a greater opportunity to breast-feed for a longer duration. Sicker women could choose to never breast-feed. Similarly, some mothers could have stopped breast-feeding when they became too ill to do so. A woman's HIV-1 disease progression or death is most likely the cause rather than the consequence of not initiating or of ceasing breast-feeding (reverse causality).16
The overall rate of mortality among mothers (32.2 per 1000 person-years), 87% of whom ever breast-fed, was similar to that reported previously from smaller maternity-based studies in which essentially all women breast-fed (eg, in Rwanda [43.8 per 1000 person-years17 and 35.0 per 1000 person-years18] and in Zaire [40.7 per 1000 person-years]19). The proportion of women who died during the 18 months after delivery in our analyses is consistent with evidence of a lower mortality rate among HIV-1-infected women within the first 2 years after delivery compared with other HIV-1-infected women20 and is intermediate between the proportions observed in 2 previous studies of mortality among HIV-1-infected mothers according to their children's feeding modality (0.88% during a mean follow-up after delivery of approximately 10 months in South Africa2 and 6% during a 24-month period of follow-up after delivery in Kenya1). Both of these studies contributed data to the pooled BHITS database.
The results of these previous 2 studies of mothers' mortality and children's feeding modality are apparently conflicting. In 1 study,13 mortality of mothers over the 24-month period after delivery was higher among those randomized to breast-feeding (18 deaths among 197 women [9%]) than among those in the formula-feeding group (6 deaths among 200 women [3%]; P = 0.009). Substantial nutritional impairment (associated with the combination of metabolic demands of breast-feeding among HIV-1-infected women who already might have borderline nutritional status and with HIV-1 infection itself) was hypothesized to have resulted in an increased risk of death. In the other study,2 no differences in mothers' mortality according to whether they chose to breast-feed their children or not were observed; over a mean follow-up period after delivery of 10 months, 0.49% (2 of 410) of women who ever breast-fed were known to have died compared with 1.92% (3 of 156) of women who never breast-fed.
The results of this individual patient data meta-analysis may help to explain the apparent contradictions between the Kenyan and South African studies. In our analyses of observed children's feeding modality and mothers' mortality, women who never breast-fed had lower median CD4+ counts than women who ever breast-fed, suggesting that women with more advanced HIV-1 disease around the time of delivery chose not to initiate breast-feeding. Women whose children were ever breast-fed could have experienced, as suggested by the Kenyan study, a “metabolic drain” over the ensuing months, such that the health of these women declined; when mortality according to never versus ever breast-fed was assessed, no statistically significant difference in mortality was detected (as in the analysis of observational data regarding children's feeding modality and mothers' mortality in South Africa). Among women who ever initiated breast-feeding, however, a longer duration of breast-feeding was associated with a lower risk of mortality, which is consistent with the concept that the longer duration itself was not protective but, instead, that those women less likely to die (healthier women) were more likely to be able to breast-feed for longer periods.
Although randomized trials are underway to evaluate exclusive breast-feeding or antiretroviral prophylaxis (maternal or infant) while breast-feeding,21 it is highly unlikely that another randomized trial of children's feeding modality per se among HIV-1-infected women will be undertaken. If not, other studies of HIV-1-infected women and their children's feeding modality should incorporate assessments of the reasons for noninitiation of breast-feeding and for cessation of breast-feeding in addition to objective measures of mothers' HIV-1 disease progression and children's health status.
Data analysis was performed by Westat (Rockville, MD) staff. Dr. Tim Farley (World Health Organization, Geneva, Switzerland) encouraged the BHITS group to undertake an analysis of HIV-1-infected mothers' mortality according to children's feeding modality.
1. Nduati R, Richardson BA, John G, et al. Impact of breastfeeding on mortality among HIV-1 infected women. Lancet
2. Coutsoudis A, Coovadia H, Pillay K, et al. Are HIV-infected women who breastfeed at increased risk of mortality? AIDS
3. The Breastfeeding and HIV International Transmission Study (BHITS) Group. Late postnatal transmission of HIV-1 in breastfed children: an individual patient data meta-analysis. J Infect Dis
4. Stewart LA, Clarke MJ. Practical methodology of meta-analyses (overviews) using updated individual patient data. Stat Med
5. Dabis F, Msellati P, Meda N, et al. 6-Month efficacy, tolerance and acceptability of a short regimen of oral zidovudine to reduce vertical transmission of HIV in breastfed children in Cote d'Ivoire and Burkina Faso: a double-blind placebo-controlled multicentre trial. Lancet
6. Jackson JB, Musoke P, Fleming T, et al. Intrapartum and neonatal single-dose nevirapine compared with zidovudine for prevention of mother-to-child transmission of HIV-1 in Kampala, Uganda: 18-month follow-up of the HIVNET 012 randomised trial. Lancet
7. The Petra Study Team. Efficacy of three short-course regimens of zidovudine and lamivudine in preventing early and late transmission of HIV-1 from mother to child in Tanzania, South Africa, and Uganda (Petra Study): a randomised, double-blind, placebo-controlled trial. Lancet
8. Wiktor SZ, Ekpini E, Karon JM, et al. Short-course zidovudine for prevention of mother-to-child transmission of HIV-1 in Abidjan, Cote d'Ivoire: a randomised trial. Lancet
9. Mandelbrot L, Msellati P, Meda N, et al. 15 Month follow up of African children following vaginal cleansing with benzalkonium chloride of their HIV infected mothers during late pregnancy and delivery. Sex Transm Infect
10. Gaillard P, Mwanyumba F, Verhofstede C, et al. Vaginal lavage with chlorhexidine during labour to reduce mother-to-child HIV transmission: clinical trial in Mombasa, Kenya. AIDS
11. Coutsoudis A, Pillay K, Spooner E, et al. Randomized trial testing the effect of vitamin A supplementation on pregnancy outcomes and early mother-to-child HIV-1 transmission in Durban, South Africa. South African Vitamin A Study Group. AIDS
12. Fawzi WW, Msamanga G, Hunter D, et al. Randomized trial of vitamin supplements in relation to vertical transmission of HIV-1 in Tanzania. J Acquir Immune Defic Syndr
13. Nduati R, John G, Mbori-Ngacha D, et al. Effect of breastfeeding and formula feeding on transmission of HIV-1: a randomized clinical trial. JAMA
14. Obermeyer CM, Castle S. Back to nature? Historical and cross-cultural perspectives on barriers to optimal breastfeeding. Med Anthropol
15. Brahmbhatt H, Gray RH. Child mortality associated with reasons for non-breastfeeding and weaning: is breastfeeding best for HIV-positive mothers? AIDS
16. Leon-Cava N, Lutter C, Ross J, et al. Quantifying the benefits of breastfeeding: a summary of the evidence. Food and Nutrition Program, Health Promotion and Protection Division, Pan American Health Organization (PAHO document reference number HPN/66/2). 2002.
17. Leroy V, Msellati P, Lepage P, et al. Four years of natural history of HIV-1 infection in African women: a prospective cohort study in Kigali (Rwanda), 1988-1993. J Acquir Immune Defic Syndr Hum Retrovirol
18. Lindan CP, Allen S, Serufilira A, et al. Predictors of mortality among HIV-infected women in Kigali, Rwanda. Ann Intern Med
19. Ryder RW, Nsuami M, Nsa W, et al. Mortality in HIV-1 seropositive women, their spouses and their newly born children during 36 months of follow-up in Kinshasa. AIDS
20. Todd J, Glynn J, Zaba B. Age patterns and trends in HIV positive and HIV negative mortality rate ratios showed clearly that postpartum women have substantially lower mortality than general infected women. Presented at: Conference on Empirical Evidence for the Demographic and Socio-economic Impact of AIDS; 2003; Durban, South Africa.
21. Gaillard P, Fowler MG, Dabis F, et al for the Ghent International AIDS Society Working Group on HIV in Women and Children. Use of antiretroviral drugs to prevent HIV-1 transmission through breast-feeding: from animal studies to randomized clinical trials. J Acquir Immune Defic Syndr
The following persons and study groups participated in the BHITS:
- Steering Committee: Francois Dabis, Valeriane Leroy, Marie-Louise Newell, and Jennifer S. Read
- Writing Committee: Anna Coutsoudis, Francois Dabis, Wafaie Fawzi, Philippe Gaillard, Geert Haverkamp, D. Robert Harris, J. Brooks Jackson, Valeriane Leroy, Nicolas Meda, Philippe Msellati, Marie-Louise Newell, Ruth Nduati, Jennifer S. Read, and Stefan Wiktor
- Data Management Center (Westat, Rockville, MD): D. Robert Harris, Susan Keast, Song Li, Rick Mitchell, Larry Muenz, Duke Owen, and Anna Rukhlya
Participating trials included the following:
- South Africa Vitamin A Study: Anna Coutsoudis, Kuben Pillay, Elizabeth Spooner, Louise Kuhn, Hoosen Coovadia, Gill Sinclair, Anne Mburu, Nolwandle Mngqundaniso, Kerry Uebel, Ingrid Coetzee, Ken Annamalai, Trevor Doorasamy, Ugene Govender, Juana Willumsen, Nigel Rollins, Jagidesa Moodley, and Daya Moodley
- Chlorhexidine Intervention Study (Mombasa, Kenya): Philippe Gaillard, Fabian Mwanyumba, Chris Verhofstede, and Marleen Temmerman
- HIVNET 012 Study: Laura Guay, Philippa Musoke, Thomas Fleming, Danstan Bagenda, Melissa Allen, Clemensia Nakabiito, Joseph Sherman, Paul Bakaki, Constance Ducar, Martina Deseyve, Lynda Emel, Mark Mirochnick, Mary Glenn Fowler, Lynne Mofenson, Paolo Miotti, Kevin Dransfield, Dorothy Bray, Francis Mmiro, J. Brooks Jackson, and Corey Duefield
- RETRO-CI Study: Stefan Wiktor, Ehounou Ekpini, John Karon, John Nkengasong, Chantal Maurice, Sibailly Severin, Thierry Roels, Moise Kouassi, Eve Lackritz, Issa-Malick Coulibaly, and Alan Greenberg
- Tanzania Micronutrient Study: W. W. Fawzi, G. I. Msamanga, D. J. Hunter, D. Spiegelman, W. Urassa, D. Mwakagile, G. Antelman, J. Coley, M. G. Herrera, W. Willett, N. McGrath, I. Ballonzi, M. Essex, B. Renjifo, S. Kapiga, M. C. Smith Fawzi, M. Garland, R. Mbise, E. Hertzmark, S. Kaaya, J. Mbwambo, C. Kagoma, and A. Swai
- ANRS049a and ANRS049b: the DITRAME (Diminution de la Transmission Mere-Enfant) ANRS 049 Study Group is organized as follows:
Epidemiology: M. Cartoux (2), F. Dabis (3, Coordinator of the DITRAME ANRS 049 trial), N. Meda (2, Coordinator of Bobo-Dioulasso Center), and P. Msellati (1, Coordinator of Abidjan Center)
Gynecology-Obstetrics: A. Bazié (2), B. Dao (2), R. Likikouet (1), L. Mandelbrot (4, Principal Investigator), and C. Welffens-Ekra (1, Principal Investigator)
Methodology: V. Leroy (3) and R. Salamon (3)
Microbiology: D. Bonard (1), P. Combe (1), M. Dosso (1), L. Gautier-Charpentier (2), F. D. Ky (2), A Ouangré (2), T. Ouassa (1), O. Sanou (2), F. Sylla-Koko (1), Y. Traore (2), and P. Van de Perre (2)
Molecular Biology: A. M. Cassel-Beraud (2), J. B. Kottan (1), O. Manigart (2), C. Montcho (1), C. Rouzioux (4), A. Simonon (2), D. Valea (2), and B. You (1)
Pediatrics: R. Camara (1), N. Elenga (1), B. Nacro (2), F. Tall (2), and M. Timité (1)
Trial Monitoring: G. Gourvellec (1), O. Ky-Zerbo (2), V. Noba (1), R. Ramon (1), I. Sombié (2), S. Tiendrebeogo (2), I. Viho (1), and S. Yaro (2)
Data Management: L. Dequae-Merchadou (3) and R. Lassalle (3)
Mother-Baby Study: Ruth Nduati, Joan K. Kreiss, Dorothy Mbori-Ngacha, Grace John Stewart, and Barbra A. Richardson
- Abidjan Center, Côte d'Ivoire
- Bobo-Dioulasso Center, Burkina Faso
- Bordeaux Coordination Unit INSERM 330, France
- Paris, France
- Writing Committee: J. Saba, G. Gray, Ch. Ndugwa, H. M. Coovadia, Ch. Kilewo, A. Massawe, P. Kituuka, P. Okong, A. Grulich, J. McIntyre, H. von Briesen, J. Goudsmit, G. Biberfeld, G. Haverkamp, G. J. Weverling, and J. M. A. Lange
- Trial Management Committee: J. M. A. Lange (Chair), J. Saba (Study Coordinator), J. McIntyre, F. Mmiro, Ch. Ndugwa, J. Moodley, H. M. Coovadia, D. Moodley, Ch. Kilewo, A. Massawe, P. Okong, P. Kituuka, H. von Briesen, J. Goudsmit, G. Biberfeld, M. Guliano, S. Declich, S. Clapp, G. Haverkamp, G. J. Weverling, D. Cooper, A. Grulich, D. Bray, F. Ngobeni, G. Baguma, S. Kyambadde, and J. Perriens
- Representatives of People Living with HIV: F. Ngobeni, G. Baguma, and S. Kyambadde
Participating Clinical Centers and Principal Investigators:
- Kampala, Uganda: Mulago Hospital (F. Mmiro and Ch. Ndugwa)
- Kampala, Uganda: Nsambya: St. Francis Hospital (P. Okong and P. Kituuka)
- Dar es Salaam, Tanzania: Muhimbili Hospital (Ch. Kilewo, A. Massawe, and F. Mhalu)
- Johannesburg, South Africa: Chris Hani Baragwanath Hospital (G. Gray and J. McIntyre)
- Durban, South Africa: King Edward Hospital (J. Moodley, H. M. Coovadia, and D. Moodley)
Trial Coordination, Central Data Management, and Statistical Analysis:
- Geneva, Switzerland: UNAIDS (J. Saba and J. Perriens)
- Amsterdam, Netherlands: Academic Medical Center, National AIDS Therapy Evaluation Center (IATEC) (J. M. A. Lange, G. Haverkamp, and G. J. Weverling)
- Frankfurt, Germany: George Speyer Haus (H. von Briesen)
- Stockholm, Sweden: Swedish Institute for Infectious Disease Control and Microbiology and Tumor Biology Center Karolinska Institute (G. Biberfeld)
- Amsterdam, Netherlands: Department of Human Retrovirology, Academic Medical Center (J. Goudsmit)
Keywords:© 2005 Lippincott Williams & Wilkins, Inc.
HIV-1; mortality; women; breast-feeding; meta-analysis