|Home | About | Journals | Submit | Contact Us | Français|
Early weaning has been recommended to reduce postnatal HIV transmission. We evaluated the safety of stopping breastfeeding at different ages for mortality of uninfected children born to HIV-infected mothers.
During a trial of early weaning, 958 HIV-infected mothers and their infants were recruited and followed from birth to 24 months in Lusaka, Zambia. Half of the cohort was randomized to wean abruptly at 4 months and the other half to continue breastfeeding. We examined associations between uninfected child mortality and actual breastfeeding duration investigating possible confounding and effect modification.
The mortality rate among 749 uninfected children was 9.4% by 12 months and 13.6% by 24 months. Weaning during the interval encouraged by the protocol (4-5 months) was associated with a 2.03-fold increased risk of mortality (95% CI: 1.13 - 3.65), weaning 6-11 months a 3.54-fold increase (95% CI: 1.68 - 7.46) and 12-18 months a 4.22-fold increase (95% CI: 1.59 - 11.24). Significant effect modification was detected such that risks associated with weaning were stronger among infants born to mothers with higher CD4 counts (>350 cells/mL).
Shortening the normal duration of breastfeeding for uninfected children born to HIV-infected mothers living in low resource settings is associated with significant increases mortality extending into the second year of life. Intensive nutritional and counseling interventions reduce, but do not eliminate, this excess mortality.
Shortening breastfeeding duration to less than 18 months was associated with significant increases in mortality among uninfected children born to HIV-infected mothers in Lusaka, Zambia. Nutritional and counseling interventions did not eliminate excess mortality associated with weaning.
HIV transmission continues to occur throughout the duration of breastfeeding prompting many advocates of prevention of mother-to-child HIV transmission programs to recommend early weaning for infants born to HIV-infected mothers [1-4]. However, benefits of breastfeeding for survival and well-being of infants and young children are well established [5,6] creating a tragic impasse for the communities, primarily in sub-Saharan Africa, most severely affected by the HIV/AIDS pandemic.
We hypothesized that early weaning at 4 months of age could be made safe enough to justify recommending this practice to prevent HIV transmission and conducted a trial to test this hypothesis in Lusaka, Zambia. Because benefits of breastfeeding are thought to decline as children age , we hypothesized that by 4 months children would have passed the critical developmental age when breastfeeding is essential to their survival. We also provided a fortified weaning cereal that needed cooking in order to minimize risks associated with preparing infant formula with contaminated water. The cereal contained all the macro- and micro-nutrient requirements of 4-6 month old infants. Cotrimoxazole was given to all infants as well as intensive counseling and education for mothers.
We have reported that the trial found no benefit of early weaning for HIV-free survival . Here we investigate the effects of weaning at different ages on mortality of uninfected children born to HIV-infected mothers. Specifically, we were interested in whether we could determine a safe age when weaning could be recommended.
We recruited a cohort of 958 HIV-infected mothers into a randomized trial of early weaning in Lusaka, Zambia. All women planned to breastfeed to at least 4 months and were counseled to breastfeed exclusively . Half of the cohort was encouraged to abruptly wean at 4 months and the other half was encouraged to continue breastfeeding and to wean at a time of their own choosing. Here we examine whether the actual age of the child when all breastfeeding ended i.e. weaning age, was associated with mortality through 24 months among uninfected children.
HIV-infected pregnant women were recruited between May 2001 and September 2004 from two antenatal clinics in Lusaka, Zambia that offered voluntary HIV counseling and testing and single-dose nevirapine prophylaxis. All women signed informed consent. The study was approved by Human Subjects Committees at the investigators’ institutions.
Blood was drawn from women at enrollment and two subsequent antenatal visits were scheduled for counseling prior to delivery. Sociodemographic and clinical history were collected at enrollment and obstetric/neonatal data after delivery. Infant heelstick blood samples were collected onto filter paper on the day of birth, at 1 week, and at 1, 2, 3, 4, 4 ½, 5, 6, 9, 12, 15, 18, 21, and 24 months of age. Clinic visits were scheduled at these same time-points and detailed questions were asked about infant feeding practices by separate study staff than those performing the counseling. Home visits were scheduled at 4 days and then again at time-points interspersed between clinic visits so that contact occurred every 2 weeks through 5 months. All randomized infants were followed on this schedule to 24 months. Home visit teams tracked participants who did not return for appointments. Information about all child deaths was sought from hospital and clinic records as well as interviews with caretakers and health care personnel. The circumstances of all deaths were reviewed and verbal autopsies conducted to identify causes of death.
Antiretroviral drugs were not available in the public sector during most of the study period but became available after May 2004 . Women eligible for treatment based on Zambian guidelines were started on first-line regimens if they consented to treatment. Cotrimoxazole was given to all women with CD4 counts <200 after November 2003 . Cotrimoxazole was given to all infants between 6 weeks and 12 months. Growth monitoring was done at least monthly, the under-five immunizations were given according to the national guidelines and children from either group with evidence of failure to thrive were provided with nutrition supplements. Children in the intervention group were provided with a 3-month supply of commercial infant formula (not pre-mixed) and a fortified weaning cereal regardless of growth status from 4 months of age. The cereal was maize meal-based fortified with milk powder, sugar, oil and micronutrients.
Maternal blood collected at enrollment was tested for CD4 and CD8 counts (FACSCount, BD Immunocytometry Systems, San Jose, CA), hemoglobin (Hemocue® system, Lake Forest, CA) and plasma viral load (Roche Amplicor® 1.5, Roche, Branchburg, NJ). Maternal CD4 counts were repeated 12 months after delivery. Infant heelstick samples were tested for HIV-1 DNA by realtime PCR at the end of the study. All positives were confirmed on at least two samples if available, and, if not, the same sample was re-tested to confirm. Amplification of the beta-globin gene was performed in all samples to ensure adequate cell numbers thereby minimizing false negative results due to inadequate sample.
Children were considered uninfected if their last available PCR test was negative. Three children who died without a prior PCR test were excluded. Uninfected child mortality through 24 months of age was analyzed using Cox Proportional Hazards models. Children were included in the analysis up to the time of their last negative test or until their death (if they died without ever testing positive). Breastfeeding cessation (weaning) was treated as a time-dependent variable. The age of weaning was calculated based on the exact age that all breastfeeding stopped. The maternal report provided at the clinic visit soonest after weaning was used. Children who died were assumed to have been breastfed up to their date of death unless study records specifically reported that breastfeeding ceased prior to the start of the illness immediately preceding the child's death. Curves to display survival probabilities by different weaning ages utilizing the time-dependent covariate for weaning were generated utilizing Cox models. Piecewise time-dependent covariates for weaning age were introduced to examine whether the effects of weaning on mortality were constant across child age. Confounding was investigated by examining whether the inclusion of each putative confounder changed the magnitude of the weaning association by more than 10%. Any factor that modified the primary relationship in this way, or which was significantly associated with mortality was included in the final model. Effect modification was investigated through stratification and multiplicative interaction terms in the Cox model. To compare characteristics between groups, differences in categorical parameters were tested using chi-squared tests, normally-distributed continuous variables using t-tests, and non-normal continuous variables using Wilcoxon rank sum tests. Analyses were done using SAS (Cary, NC).
Of 958 mother-child pairs randomized in the study, 749 had a last PCR test which was negative. The mortality rate among uninfected children was 2.8% (21 deaths) by 4 months, 9.4% (65 deaths) by 12 months and 13.6% (91 deaths) by 24 months. The median time between the last negative result and death was 31 days (inter-quartile range [IQR] 16 - 53). The early mortality rate in the randomized cohort is an under-estimate because the study aimed only to randomize mother-child pairs surviving to one month of age. In order to generate estimates of under two year mortality that do not under-represent neonatal mortality, we examined neonatal mortality in the larger cohort of 1002 live-births, including those excluded prior to randomization and censoring those who tested HIV positive before 28 days. In this birth cohort, neonatal mortality was 4.4%.
Maternal, household and infant characteristics of the surviving and deceased uninfected children are displayed in Table 1. Children who died were more likely to have mothers with low CD4 counts and high viral loads, were more likely to have had mothers who died, to have lived in households reporting food insecurity and with more than one child under 5 years of age, to have been of low birth weight and to have weaned earlier.
Diarrhea was the leading cause of death contributing to 61.5% of deaths 6-24 months but was not a major cause of mortality for those under 6 months of age. Pneumonia was the second major cause of death contributing to deaths across all age categories (42.9% of deaths overall). If sepsis and pneumonia are combined into one category, they became the leading cause of death contributing to 84.6% of deaths under 6 months and to 41.5% of deaths 6-24 months (Table 2).
Weaning was associated with a more than 2-fold increase (Hazard Ratio [HR]=2.56 95% CI: 1.53-4.28) in mortality over the full period from birth to 24 months. Under two-year mortality in the cohort of children who did not stop breastfeeding before 18 months was 9.7% and was 17.4% in the cohort who stopped breastfeeding at 5 months (Figure 1).
To investigate if weaning became safer as children became older, we examined the effects of weaning before 4 months, 4-5 months, 6-11 months and 12-18 months utilizing four time-dependent covariates for these periods with those who weaned at ages older than 18 months as the comparison. Elevated risks of uninfected child mortality persisted through 18 months. Weaning at 4-5 months, the period encouraged in the intervention group and especially supported with additional nutritional interventions, was associated with a significant 2.03-fold elevation in mortality (95% CI: 1.13 - 3.65), albeit slightly lower than observed in other intervals. Weaning 6-11 months was associated with a 3.54-fold increase (95% CI: 1.68 - 7.46) and weaning 12-18 months a 4.22-fold increase (95% CI: 1.59 - 11.24) in uninfected child mortality compared to those weaning at ages older than 18 months (Table 3). The adverse effects of weaning were not confined to the period immediately after breastfeeding ended. Among children who stopped breastfeeding at 4 months, the absence of breast milk was associated with significantly increased mortality in the period 2 months after weaning (RH=2.17 95% CI: 1.13-4.16). In the acute period (within the first 2 months after weaning), weaning-related mortality was not as elevated (RH=1.61 95% CI: 0.48-5.42).
We investigated if any other factors either alone or together confounded the relationship between weaning and mortality. Low maternal CD4 count during pregnancy, maternal mortality, low birth weight, having more than one other child in the household less than 5 years old and reported food insecurity were each independently associated with significantly increased uninfected child mortality in multivariate analysis but did not account for the relationship between weaning and mortality. After adjustment for these factors, the absence of breastfeeding continued to be associated with significantly increased risk of uninfected child mortality (Table 4). Weaning was associated with significantly increased risk of mortality in both the intervention group (RH=3.19 95% CI: 1.15-8.83) and the control group (RH=2.84 95% CI: 1.38-5.83) after adjusting for these same factors.
Maternal CD4 count measured during pregnancy was a significant effect modifier of the relationship between weaning and mortality. When maternal CD4 counts were high (>350 cells/ml), weaning was associated with a more than 4-fold increased risk of death (HR= 4.16 95% CI: 1.90 – 9.13) after adjusting for other confounders (Table 4). The weaning-mortality association was attenuated when maternal CD4 counts were low. Correspondingly, the association between low maternal CD4 counts and increased uninfected child mortality was strongest among breastfed infants (Figure 2, Table 4). Taking into account this interaction, as well as adjusting for maternal mortality, birth weight, other children in the household and food insecurity, the relative hazard of death associated with weaning before 6 months was 3.64 (95% CI: 1.63-8.14), weaning 6-11 months 7.63 (95% CI: 2.84-20.45), and weaning 12-18 months 8.96 (95% CI: 2.94-27.29).
There was no evidence that markers of more beneficial socioeconomic position (including availability of electricity, domestic water source, maternal education, cooking facilities etc.) modified the association between weaning and mortality. None of the interaction terms was close to statistical significance. Further adjustment for maternal initiation of antiretroviral therapy or for the era when antiretroviral therapy became available had no appreciable effect on the results.
Premature truncation of the usual breastfeeding duration more than doubled the risk of death among uninfected children and the increase was more than 4-fold if mothers were not yet immunocompromised. Weaning at older ages continued to confer elevated mortality risks with 3- and 4-fold elevations with weaning between 6-11 months and 12-18 months, respectively, relative to breastfeeding for longer than 18 months. Mortality elevations were > 7-fold higher with weaning 6-11 months and 12-18 months if mothers had CD4 counts >350 cells/uL. These sobering results are consistent with prior studies which have observed among the general population that breastfeeding continues to protect against mortality even into the second year of life [5,7,12].
There was some evidence that the study interventions to support early weaning were beneficial. These interventions included counseling, education about preparation of replacement feeds and hygiene, growth monitoring, including nutritional supplementation of children manifesting signs of failure to thrive, cotrimoxazole, and provision of a 3-month supply of nutritionally-replete replacement foods. Hazard ratios associated with weaning during the study-supported interval of 4-5 months were slightly lower than elevations observed at older ages. Despite these interventions, weaning was associated with a more than 2-fold higher risk of death than continued breastfeeding. It is unclear what more could be added to make weaning safer than what was done as part of our study. It is possible that we have reached a biological threshold. Moreover, adverse effects of early weaning were not confined to the acute period immediately after weaning, suggesting that it is the absence of breast milk, rather than some temporary difficulty of the child adjusting to change, that is responsible for elevated mortality. It is likely that death due to weaning will be greater in real-world program situations where the extent of education and nutritional support is less than provided by our trial. Higher mortality rates have been observed in program settings .
We did not observe attenuation of the hazard ratios associated with weaning among women with more advantaged socioeconomic characteristics but numbers of women who met even crude indicators of advantage were small. Benefits of breastfeeding are demonstrated even in well-resourced settings e.g. in the United Kingdom, where breastfeeding was found to be associated with significantly reduced risks of severe diarrhea and pneumonia-related morbidity resulting in hospitalization . Our hazard ratios may be biased towards the null by the nature of clinical research which over-represents motivated and compliant participants and provides a health service safety net with easier access to medical care and medications.
It is intriguing that the benefits of breastfeeding were greatest among women with higher CD4 counts. Although some risks associated with weaning may be in part related to environmental factors, the effect modification by maternal immune status highlights the importance of immunologically-active components of breast milk [15,16]. Our findings suggest that there may be some deficiencies in breast milk of immunocompromised women but these are yet to be identified .
Our results pertaining to the effects of early weaning on uninfected child mortality should be viewed in the context of HIV-free survival. The only reason to encourage early weaning is to reduce HIV transmission. Guidelines for uninfected women are unambiguous in their support for breastfeeding to 24 months or longer. Our study was conducted largely prior to the availability of antiretroviral therapy for women. We have previously reported in both intent-to-treat analysis and in analyses based on actual behavior that early weaning resulted in no net benefit for HIV-free survival. Benefits for HIV reduction were off-set by elevations in uninfected child mortality. There are encouraging new data that antiretroviral drug regimens given as prophylaxis to the infant can reduce post-natal HIV transmission [20,21]. Mothers receiving effective antiretroviral therapeutic regimens also appear to be at low risk of HIV transmission [22,23]. When antiretroviral drugs are given, the risks for uninfected child mortality take on greater salience as even small elevations can counter-balance the now lessened HIV transmission risks. The balance also shifts among women who are at low risk of transmitting such as women with high CD4 counts. As we demonstrate here, women least likely to transmit HIV because of higher CD4 counts are also those for whom stopping breastfeeding confers the greatest dangers. As we have previously reported, even in the absence of antiretroviral therapy in women with higher CD4 counts, there is net benefit in terms of HIV-free survival with longer breastfeeding .
There are limitations of our analysis. We aimed to reduce the risks of reverse causality by reviewing the clinical circumstances of each death to exclude deaths where underlying illness was the motivation for weaning but we acknowledge that we may not have had all necessary information in every case. Our observation that early weaning was associated with death in both the intervention and the control group, despite the vastly different reasons for early weaning in these two groups, strengthens our inference about the effects of early weaning on mortality. We also investigated possible confounding by several socioeconomic and clinical factors known to be associated infant and young child death and none of these factors accounted for the associations. Unmeasured confounders may play a role but, given the consistency of our results with biological plausibility and studies in uninfected populations, this is unlikely. One of the greatest strengths of our cohort is the degree of heterogeneity of feeding practice. Other studies among HIV-infected women tend to be more homogenous in their feeding practices limiting the comparisons possible. The fact that women in the intervention group were specifically encouraged to wean early also minimizes the usual reasons for early weaning making our associations less prone to the confounding factors that may dominate in other studies precluding sufficient variability for adequate statistical adjustment.
Support is needed for programs in low resource settings to incorporate antiretroviral therapy for pregnant women with low CD4 counts and other strategies addressing HIV transmission over the postnatal period so that breastfeeding for a normal duration can be unambiguously supported. Our data, consistent with prior data among uninfected children, demonstrate that survival of uninfected children born to HIV-infected mothers is compromised through the second year of life when breastfeeding is stopped early. Nutrition, education and counseling interventions may reduce, but do not eliminate, this excess mortality.
We would like to thank the Zambian families who participated in the research and all the study staff and volunteers. We gratefully acknowledge assistance with aspects of the design and conduct of the study from: Drs. Marc Bulterys, Elwyn Chomba, Lynne Mofenson, Ellen Piwoz, Kevin Ryan, Nancy Scott, Cheswa Vwalika, Jan Walter.
The study was supported by grants from the National Institute of Child Health and Human Development (NICHD), National Institutes of Health (NIH) (R01 HD 39611 and R01 HD 40777). GMA is a recipient of the Elizabeth Glaser Pediatric AIDS Foundation (EGPAF) Scientist Award. None of the authors has any conflicts of interest to declare.