Introduction
The global HIV epidemic has had a major impact on the health and survival of infants and children in sub-Saharan Africa [1-3]. The rate of HIV transmission via breast-milk ranges from 12 to 26%, depending on the duration of breastfeeding, the recent timing of seroconversion and the presence of mastitis [4], and appropriate guidelines for counselling HIV-infected mothers on breastfeeding are a critical public health issue. Non-breastfeeding and weaning has been associated with increased infant and child morbidity and mortality [5-8], but previous studies did not take into account the reasons mothers failed to initiate breastfeeding or chose to wean their children. Prolonged breastfeeding is almost universal in sub-Saharan Africa, particularly in rural areas [9]. In cultures in which lactation is normative, it is likely that the decision not to start breastfeeding or to wean early is involuntary, and probably reflects an inability to breastfeed, either because of maternal illness, or infant morbidity reducing the capacity to suckle. For example, a study in Peru and the Dominican Republic [10] showed that low birthweight infants were 60% more likely never to be breastfed than infants of normal birthweight. Estimates of infant or child mortality associated with non-breastfeeding that ignore the reasons for the non-initiation of lactation, may thus overestimate the mortality risk that might be associated with volitional behaviors. Similarly, weaning has been associated with increased mortality [5-8], but there is a problem of determining causality, because weaning may be precipitated by pre-existing maternal or infant illness, rather than a voluntary maternal decision. The risk of mortality associated with weaning may thus be overestimated if preceding morbidity in the mother or child is ignored [8].
Mortality after the voluntary non-initiation of breastfeeding or voluntary weaning can provide an appropriate benchmark with which to assess the potential risks and benefits of non-breastfeeding or weaning at 3-6 months, as recommended by current UNICEF/WHO guidelines to avoid the transmission of HIV via breast-milk [11]. We analysed Demographic and Health Survey (DHS) data to estimate the risk of mortality among children who were never breastfed or were weaned because of preceding morbidity in the mother or infant, compared with mortality among children weaned for voluntary (i.e. for non-health-related) reasons, in order to assess the risks to children, if mothers adhere to UNICEF/WHO guidelines.
Methods
Populations and data
For more than 15 years, DHS has been developing its reputation as a leader in the field of demographic and health research by collecting comparative data on a global scale. DHS [12] are nationally representative household surveys of random samples of 4000-8000 eligible female respondents. The standard DHS collects information on family planning, maternal and child health, child survival, AIDS, educational attainment, and household composition and characteristics. The methods used to conduct DHS surveys in different countries are standardized, with all similar questions asked in different countries recoded to have the same variable name. The aim is to make comparison of the same variables across different countries feasible, and the pooling of data for comparison across countries is facilitated with use of variables given the same code and weights given to each country based on the sample size available. The DHS questionnaire also obtains information on breastfeeding of the most recent child, including questions on whether breastfeeding was ever initiated, whether the mother had weaned by the time of the interview, and why the mother had not initiated breastfeeding or had stopped breastfeeding [12]. Weaning was defined as having stopped breastfeeding completely for women who had initiated breastfeeding. The specific questions asked were 'why did you never breastfeed (NAME) at the breast?' and 'why did you stop breastfeeding (NAME)?'. The woman answered the question based on a list of responses given to her, and we categorized these responses into 'voluntary choice' or 'preceding morbidity'. Voluntary choice included women who had not initiated or had stopped breastfeeding because the child was of appropriate weaning age, women who had started working, or who became pregnant again, women who felt they had insufficient milk, or started using a family planning method. The response 'insufficient milk' was included in the voluntary choice category because other studies, including a WHO study, found that a substantial percentage of women (28-59% in countries studied) reported insufficient milk as a reason for not breastfeeding because of a perception of there not being enough milk to fulfill the child's nutritional needs [13]. Preceding morbidity included mothers who stated that they did not initiate breastfeeding or weaned because they were ill, had nipple problems that interfered with feeding, or that the child was ill and refused or was incapable of breastfeeding. If the child died before the initiation of breastfeeding or weaning they were excluded from the analyses. The information on the most recent child was used to assess survival among never-breastfed infants, and children who were weaned by reasons for non-breastfeeding or weaning. Efforts were made to obtain accurate information on the exact date of birth and birthweight by probing carefully and utilizing information on health cards whenever possible.
The most recent DHS survey data from countries in sub-Saharan Africa, Asia, Latin American and the Caribbean were used. The countries selected were Uganda (1995), Kenya (1998), Zimbabwe (1994), Zambia (1996), Ghana (1998), Pakistan (1991), Bangladesh (1997), Haiti (1994), Morocco (1992), Dominican Republic (1996), Paraguay (1990), Brazil (1996), Philippines (1998) and Colombia (1995). All 14 countries provided data on the reasons for weaning and 10 countries had information on the reasons for not initiating lactation.
Statistical analysis
Mother-infant pairs were divided into those who never breastfed and women who had breastfed in the past, and both populations were stratified by the reason mothers gave for why they never initiated breastfeeding, or weaned (i.e. voluntarily or as a result of preceding morbidity). Child mortality was calculated as a rate per 1000 children at risk within these two strata. Children currently breastfeeding at the time of the interview were not included in this analysis. Rate ratios (RR) of mortality were estimated for children who were never breastfed or who were weaned because of health reasons, relative to mortality among children never breastfed or weaned for voluntary reasons (referent group). Kaplan-Meir survival analysis was used to estimate survival curves, truncated at 24 months of age, for the never-breastfed and weaned, by the reason for the non-initiation of breastfeeding or weaning. Log-rank chi square tests were used to test for equality of the survival curves across strata, and hazard ratios were estimated for cumulative mortality at 24-months of age [14]. Country-specific data were aggregated to generate combined mortality rates and survival. To estimate the bias in child mortality if the reasons for non-initiation or weaning are ignored, we determined the mortality rates of children not breastfed or weaned for all reasons, relative to those not breastfed or weaned for voluntary reasons. To assess whether previous maternal or infant morbidity might be associated with the non-initiation of lactation, we examined the proportions of infants who were never breastfed among mothers who reported complications of labor and delivery (e.g. prolonged labor, hemorrhage, convulsions), versus mothers without complications and among infants of low birthweight (< 2500 g) compared with normal birthweight infants. To assess the effect of normative lactation practices, we also examined the differences between countries with shorter (≤ 18 months) and longer (> 18 months) median durations of breast-feeding.
Results
Mortality by reasons for not breastfeeding and weaning
Child mortality for never-breastfed children is shown in Table 1. The overall mortality for children never breastfed for all reasons was 221.3 per 1000 (387/1749). Preceding maternal-infant morbidity was reported as a reason for not breastfeeding by 63.9% of mothers (1117/1749), and child mortality associated with never initiating breastfeeding because of preceding morbidity was significantly higher than the non-initiation of breastfeeding as a result of voluntary choice [326.8 per 1000 versus 34.8 per 1000, respectively, Mantel-Haenszel weighted RR 7.61, 95% confidence interval (CI) 5.03-11.52]. The overall estimate of child mortality associated with the non-initiation of breastfeeding (221.3 per 1000) is overestimated by as much as 536% (i.e. 221.3/34.8) by including in the estimate children never breastfed because of preceding illness.
Child mortality rates and the rate ratios of mortality associated with involuntary weaning relative to voluntary weaning are shown in Table 2. Mortality was consistently higher among children weaned because of preceding morbidity compared with those weaned for non-health-related reasons (19.2 per 1000 versus 9.3 per 1000, respectively, Mantel-Haenszel weighted RR 2.10, 95% CI 1.66-2.66). The overall mortality among children weaned for all reasons was 12.2 per 1000 (270/22 124), which was 31.2% higher than mortality associated with voluntary weaning (i.e. 12.2/9.3).
Child mortality among the offspring of women who never initiated breastfeeding was significantly higher than among the children of women who weaned (RR 8.1, 95% CI 15.6-21.0). The risk was higher both among children who were never breastfed versus children who were weaned because of preceding morbidity (RR 17.0, 95% CI 14.0-20.6), and among children who were never breastfed versus those who were weaned as a result of voluntary choice (RR 3.72, 95% CI 2.4-5.8).
Fig. 1 shows 24-month Kaplan-Meir survival curves of children who were never breastfed, stratified by the maternal reason for never breastfeeding. There was a significantly lower survival of children never breastfed because of preceding morbidity compared with children never breastfed as a result of maternal choice (log rank χ2 151.6; P < 0.0001). The cumulative mortality at 24 months was 24.0% for children never breastfed because of preceding morbidity, compared with 4.0% for children never breastfed for non-health-related reasons (hazard ratio of child death 7.30, 95% CI 5.03-10.60). Country-specific differentials in survival between infants not breastfed for health reasons versus maternal choice were significant in all countries except Ghana, Zimbabwe and Haiti (Table 1).
Fig. 2 shows 24-month survival curves for the breastfed children, stratified by the reason for weaning. There was a significantly lower survival of children weaned because of preceding morbidity compared with children weaned voluntarily (log rank χ2 46.0; P < 0.0001), and the cumulative mortality at 24 months was 1.1% for children weaned because of preceding morbidity, compared with 0.38% for children weaned for voluntary reasons (hazard ratio of child death 2.24, 95% CI 1.76-2.85). Kaplain-Meir survival curves for each country were also examined, and there was a consistently higher mortality among children weaned because of preceding morbidity compared with children weaned as a result of voluntary maternal choice, and these curves were significantly different in all countries except Uganda, Morocco, Philippines, Brazil and Paraguay (Table 2).
Evidence of self-selection by normative breastfeeding patterns in populations with long and shorter median durations of breastfeeding
The non-initiation of breastfeeding was less common in populations with prolonged lactation (2.4%, 626/25 599), in which the median duration of breastfeeding was more than 18 months, compared with populations with shorter breastfeeding durations (6.9%, 1482/21 421; P < 0.0001). Also, in the longer breastfeeding populations, out of 323 women who did not breastfeed, 82.7% reported maternal or infant illness as the reason for the non-initiation, whereas in the shorter breastfeeding populations, among the 1426 women who never breastfed, 59.6% cited maternal or infant health as their reason (P < 0.0001). Therefore, in societies in which prolonged breastfeeding is the norm, women are less likely not to breastfeed, and if they fail to initiate lactation, they are more likely to do so because of preceding morbidity, compared with societies with shorter median breastfeeding durations.
The non-initiation of breastfeeding was more common among women reporting obstetric complications. In the longer breastfeeding populations, among 7975 mothers reporting any delivery complications, 2.2% did not initiate breastfeeding, whereas in 10 478 mothers without reported complications, 1.3% never breastfed (P < 0.001). In the shorter breastfeeding populations, among 2552 mothers who reported any delivery complications, 9.5% did not initiate breastfeeding, and among 8009 mothers without obstetric complications, 8.0% never breastfed (P < 0.01). Non-breastfeeding was more common among low birthweight infants. In the longer breastfeeding populations, 3.2% of infants who were low birthweight were never breastfed compared with 1.8% of normal birthweight infants (P < 0.001). In the shorter breastfeeding populations, 11.2% of low birthweight infants were not breastfed, whereas among normal birthweight babies, 7.3% were never breastfed (P < 0.001).
The mortality of non-breastfed children was 516 per 1000 in the longer breastfed populations (323/626), compared with 173 per 1000 in the shorter breastfeeding populations (257/1482; P < 0.001). This further supports the evidence that the failure to initiate breastfeeding is more selectively caused by preceding morbidity in societies in which universal and prolonged breastfeeding is the norm. The mortality associated with weaning caused by preceding morbidity was also higher in the long breastfeeding populations (310 per 1000, 72/2299), than the shorter breastfeeding populations (120 per 1000, 51/4096; P < 0.001), again suggesting greater self-selection in the former group of countries.
Discussion
There are clear and well-documented benefits of breastfeeding for child nutrition, health, growth and survival [5-7]. This paper focuses on mortality associated with reasons for the non-initiation of breastfeeding or weaning, and the potential biases of self-selection and reverse causality, which may affect estimates of risk. In all the 14 developing countries examined, child mortality among children not breastfed or weaned because of preceding morbidity in the mother or infant, was higher than mortality among children not breastfed or weaned as a result of maternal choice (Table 1 and Table 2). The cumulative hazards of child mortality at 24 months were over seven times higher among infants who were never breastfed because of preceding morbidity, compared with those not breastfed by maternal choice (Fig. 1), and over two times higher among children weaned because of preceding morbidity, compared with children who were weaned for non-health-related reasons (Fig. 2).
The non-initiation of breastfeeding was much less common in the longer breastfeeding populations (2.4%), compared with the shorter breastfeeding countries (6.9%), suggesting greater self-selection in the former populations. It is noteworthy that 63.9% of the mothers who did not initiate breastfeeding did so because of preceding morbidity, and this reason for non-breastfeeding was more frequently cited by mothers in long breastfeeding cultures (82.7%) than shorter breastfeeding cultures (59.6%). Mortality among non-breastfed children was substantially higher in longer breastfeeding populations (516 per 1000), compared with shorter breastfeeding countries (173 per 1000). Mortality associated with weaning as a result of preceding morbidity was also higher in the longer breastfeeding countries compared with the shorter breastfeeding countries (310 per 1000 and 120 per 1000, respectively). This again illustrates that in cultures in which breastfeeding is common and long, the non-initiation of breastfeeding or weaning early is more likely caused by a preceding illness. Also, women who reported delivery complications or a low birthweight infant were significantly more likely not to initiate breastfeeding, again suggesting selective effects. This is consistent with another study [10], which found that a higher proportion of mothers failed to initiate lactation if their infants were low birthweight. It is also possible that countries with longer durations of breastfeeding also represent populations with poorer access to health services, leading to more severe morbidity, and subsequently a more pronounced effect of illness on the decision to breastfeed.
Our findings with respect to mortality are consistent with other reports. A Malaysian study [8] found that after excluding infants who were too ill to breastfeed, the excess mortality associated with non-breastfeeding relative to breastfed infants, was markedly reduced and was no longer statistically significant. Also, a randomized trial of breastfeeding versus formula feeding among HIV-infected women in Kenya [15], found comparable mortality rates among formula-fed and breastfed infants. A WHO collaborative study [7] attempted to control for preceding morbidity by assessing the breastfeeding status before the onset of the terminal illness or 7 days before death. However, in Pakistan, where breastfeeding was almost universal, the odds ratio of mortality associated with non-breastfeeding or weaning during the first 2-3 months of life was two- to fourfold higher than the odds of mortality in Brazil and the Philippines, where a lower proportion of women initiated lactation or breastfed for at least 2-3 months [7]. This suggests self-selection and reverse causality among the minority of Pakistani mothers who failed to initiate breastfeeding, or who weaned early. In Guinea-Bissau, Jakobsen et al. [16] also found that earlier weaning was associated with preceding illness of the child or mother.
There are limitations to our study. We cannot assess the reliability of self-reported information on the reasons for weaning. However, our findings with respect to child mortality risks among the never-breastfed or children weaned because of preceding morbidity, relative to those voluntarily weaned, were observed in all the countries examined, which indicates a strong and consistent association. However, future research is needed on reasons for the non-initiation of breastfeeding or weaning in order to validate maternal self-reports. Also, although birthweight information was obtained from health cards whenever possible, when this was not available, there could be problems with recall. However, results of weight-by-age for the countries used are published in DHS manuals, and the presence of minimal heaping shows that anthropometric data are not influenced by the misreporting of ages.
Application to UNICEF/UNAIDS guidelines on breastfeeding by HIV-positive mothers
Mortality among the voluntarily non-breastfed or weaned children represents the best estimate of the potential risk if HIV-positive mothers decide not to initiate breastfeeding, or to stop breastfeeding in order to prevent breast-milk HIV transmission [12]. However, previous studies that did not account for the reasons mothers do not initiate breastfeeding, or why they stopped breastfeeding [5-8,10] are likely to overestimate the adverse effects of the non-initiation of breastfeeding or weaning, by including mortality among children weaned because of pre-existing morbidity. Relative to mortality among infants not breastfed or weaned by maternal choice, we found that the total mortality was overestimated by 536% for non-initiation and by 31.2% for weaning, because of the inclusion of children not breastfed or weaned as a result of preceding morbidity. There is debate about whether to recommend that HIV-positive mothers initiate breastfeeding or to wean. To the best of our knowledge, this is the first study to account for self-selection and reverse causality using unique information on the mothers' stated reasons for the non-initiation of breastfeeding and weaning. These results underscore the importance of the reasons for not breastfeeding and weaning and the consideration of normative feeding practices, when estimating child mortality associated with the volitional non-initiation or cessation of breastfeeding. Such considerations should enter into policy recommendations, such as the UNICEF/UNAIDS guidelines for breastfeeding by HIV-infected mothers.
The objective of the UNICEF/UNAIDS guidelines is to maximize HIV-free survival time by minimizing exposure to HIV infection via breast-milk. The average mother-to-child transmission rate via breastfeeding is approximately 16.0% [14]. The mortality rates associated with voluntary non-breastfeeding or weaning from our study are 3.5 and 0.9%, respectively (Table 1 and Table 2). Therefore, if infants of HIV-positive mothers are breastfed we would expect 16.0% to become HIV infected and 0.9% to die (16.9%), giving a net HIV-free survival of 83.1%. However, if HIV-positive mothers voluntarily decide not to breastfeed, the expected mortality would be 3.5%, and the net HIV-free survival would be approximately 96.3%. Therefore, there is a net HIV-free survival benefit of 13.2% associated with the voluntary non-initiation of breastfeeding. When mother-to-child HIV transmission rates via breastfeeding are accounted for, the net HIV-free survival is greater for non-breastfeeding infants, and exceeds that of breastfed infants who are then weaned. It is also important to consider that present estimates are based on the current practices of feeding, and efforts to provide safer alternatives to breastfeeding for HIV-infected mothers will result in even more significant reductions in child mortality associated with non-breastfeeding. Similarly, the risk of mother-to-child transmission associated with early breastfeeding is likely to be underestimated because highly infectious mothers are likely to infect earlier.
The counselling of HIV-positive women on the risks of breastfeeding or substitute feeding should incorporate appropriate estimates of mortality risk and, as shown by this analysis, previous estimates of mortality associated with non-initiation or weaning are likely to be inflated by uncontrolled confounding as a result of pre-existing morbidity. The implications of these findings are that voluntary non-breastfeeding or weaning may be less hazardous than suggested by previous studies, and that the UNICEF/WHO recommendations to refrain from breastfeeding may pose a lower risk to the health of infants than was hitherto believed. Moreover, as suggested by the present analyses, policies on breastfeeding or weaning by HIV-positive mothers may need to be tailored to normative breastfeeding practices, as reflected by median durations of breastfeeding in different countries.
In conclusion, child mortality as a result of the voluntary non-initiation of breastfeeding or voluntary weaning is lower than previously estimated, after controlling for the reverse causality due to preceding morbidity in the mother or child. Mortality in infants whose mothers voluntarily chose not to initiate breastfeeding, or decided to wean their children, should be used as benchmarks for counselling on breastfeeding by HIV-positive mothers.
Acknowledgements
The authors wish to acknowledge the generous support of Fogarty International Center grants nos. 2D43 TW000010, 5D43TW00010, and 5P30HD06826, National Institute of Child Health and Development, and grants from William and Flora Hewlett grant no. 98-2409. The authors are also grateful to Demographic and Health Surveys, Macro International Inc. for permission to use these data sets. They would also like to thank Drs Taha E. Taha, Lynn Mofenson and Laura Porter for their comments.
Sponsorship: This work was supported by Fogarty grant no. 2D43 TW000010 and William and Flora Hewlett grant no. 98-2409.
References
1.Fowler MG. Update: transmission of HIV-1 from mother to child. Curr Opin Obstet Gynecol 1997, 9:343-348.
2.UNAIDS. Report on the global HIV/AIDS epidemic. Geneva: UNAIDS; 2000.
3.Preble EA. Impact of HIV/AIDS on African children. Soc Sci Med 1990, 31:671-680.
4.Leroy V, Newell ML, Dabis F, Peckham C, Van de PP, Bulterys M, et al. International multicentre pooled analysis of late postnatal mother-to-child transmission of HIV-1 infection. Ghent International Working Group on Mother-to-Child Transmission of HIV [published erratum appears in Lancet 1998; 352:1154]. Lancet 1998, 352:597-600.
5.Briend A, Wojtyniak B, Rowland MG. Breast feeding, nutritional state, and child survival in rural Bangladesh. BMJ (Clin Res Ed) 1988, 296:879-882.
6.Cunningham AS, Jelliffe DB, Jelliffe EF. Breast-feeding and health in the 1980s: a global epidemiologic review. J Pediatr 1991, 118:659-666.
7.WHO Collaborative Study Team on the Role of Breastfeesing on the Prevention of Infant Mortality. Effect of breastfeeding on infant and child mortality due to infectious diseases in less developed countries: a pooled analysis. Lancet 2000, 355: 451-455.
8.Habicht JP, DaVanzo J, Butz WP. Does breastfeeding really save lives, or are apparent benefits due to biases? Am J Epidemiol 1986, 123:279-290.
9.Haggerty P, Rutstein SO. Breastfeeding and complementary infant feeding, and the postpartum effects of breastfeeding. Demographic and Health Surveys, Comparative Studies No.30. Calverton, MD: Macro International Inc.; 1999.
10.Elo IT, Miller JE. Ever-breastfed status and health at birth. In: Measurement of maternal and child mortality, morbidity and health care; interdisciplinary approaches. Boerma JT (editor). Belgium: International Union for the Scientific Study of Population; 1992. pp. 309-331.
11.UNAIDS. WHO, UNICEF, UNAIDS statement on current status of WHO/UNAIDS/UNICEF policy guidelines. Geneva: UNAIDS; 1999.
12.Macro International Inc. Demographic and Health Surveys. Calverton, MD, USA: Macro International Inc.; 2002.
13.World Health Organization. Contemporary patterns of breastfeeding: report on the WHO collaborative study on breast-feeding. Geneva: WHO; 1981.
14.Altman DG. Practical statistics for medical research, 8th ed. London: Chapman and Hall; 1997.
15.Nduati RW, John GC, Mbori-Ngacha D, Richardson BA, Overbaugh J, Mwatha A, et al. Effect of breastfeeding and formula feeding on transmission of HIV-1. JAMA 2000, 283:1167-1174.
16.Jakobsen MS, Dodemann M, Molbak K, Aaby P. Reason for termination of breastfeeding and the length of breastfeeding. Int J Epidemiol 1966, 25:115-121.
© 2003 Lippincott Williams & Wilkins, Inc.