The findings from this analysis show strong associations between NBF and morbidity at each age interval. Rates of illnesses and/or admissions and malnutrition were consistently higher in the NBF group than in the BF group; the adjusted rate ratios for illnesses and/or hospital admission were significantly higher in the NBF group than in the BF group; similarly, the adjusted rate ratios of malnutrition were consistently higher in the NBF group than in the BF group. These findings support our original hypothesis and suggest that the adverse effects of weaning occur quickly and manifest as acute events resulting in clinic visit or hospital admission. These complications occurred despite continuous counseling about BF and provision of care for women and children in the PEPI-Malawi study .
We also estimated cumulative mortality at 6–15 months of age. The cumulative mortality rate at 15 months in the NBF group was approximately twice the rate in the BF group (6.4% vs 3.5%). However, these rates were similar during 6–12 months of age and substantially higher in the NBF group during 12–15 months of age. Although there were no deaths after 12 months in the BF group, it should be noted that the denominator becomes small after 12 months (). After adjusting for other factors, NBF was significantly associated with higher hazard of death. The association of cessation of NBF with mortality appears to be substantially higher than the baseline rate among BF infants starting in the second year. These results agree with findings from another study that showed that early cessation of BF is associated with increased mortality that extends to the second year among HIV-uninfected African children born to HIV-infected mothers [17
]. Other studies that examined interval-specific effect of BF were mainly focused on the effect of early exclusive BF on morbidity .
An important secondary finding in this study was that TMP-SMX prophylaxis to prevent PCP of HIV-exposed, uninfected children was significantly associated with lower frequency of illnesses and/or admissions at each interval (and similarly early malnutrition) after simultaneously controlling for BF and other factors. PCP prophylaxis was provided to HIV-exposed BF infants as part of clinical care and was not part of the main PEPI trial in Malawi. Although this finding is observational and should be interpreted with caution, it reveals that TMP-SMX has benefits and potentially could prevent other infections that could lead to illnesses and hospital admissions among HIV-uninfected children. Similar benefits of reductions in hospital admissions, antibiotic prescriptions, and death have been reported from other countries in Africa among HIV-infected children [19
]. In addition, TMP-SMX taken by HIV-infected adults was shown to reduce morbidity and mortality among other HIV-uninfected family members [21
]. In Zambia, a study among HIV-infected children reported reductions in morbidity and mortality despite high background TMP-SMX resistance [22
]. A clinical trial (the PROMISE study) is planned to start soon to assess whether extended TMP-SMX prophylaxis can reduce morbidity and mortality after BF cessation among HIV-exposed, uninfected children (clinical trial NCT01061151).
A potential bias in our study is censoring of infants who became HIV infected during the study follow-up period, because most of these infants were in the BF group. Therefore, it is likely that the BF group would appear to be healthier than the NBF group, and some of the differences in morbidity and mortality may have been over-estimated. However, the rates of HIV infection during 6–15 months of age were low. Another limitation common to BF studies is how to address the issue of reverse causality [23
] (ie, the association between NBF and increased infant morbidity and mortality is caused by other factors [eg, infant illness] that may have lead to cessation of BF or switching the child to NBF). Therefore, the outcome of illness or death is the cause of NBF instead of being the result of NBF. There is no general agreement on how to address this potential bias. In our study, some of the analytic approaches and comparisons that we performed suggest that the effect of this bias could be minimal. In the morbidity analyses, we examined the association of disease frequency and malnutrition with BF and NBF at 3-month intervals. The results at each interval were consistent after adjusting for other risk factors. Excluding from the analysis infants of women who switched during the interval from BF to NBF did not change the differences in morbidity rates. For the mortality analysis, we used an extended Kaplan–Meier analysis approach, which addressed the exposure to BF as a time-varying covariate when assessing its association with mortality. In the extended Kaplan–Meier curves, participants contributed to different curves, representing hypothetical cohorts at different times during follow-up [16
]. We compared the characteristics of women at each age interval whether they remained BF or NBF (). The characteristics of the BF and NBF women remained stable over the study duration, with almost no change between intervals, suggesting that there were no major changes in underlying characteristics of women or children at each age interval because of excessive movement or mixing between the groups of BF and NBF women. Of note, there were more women with low CD4 cell count (<250 cells/mm3
) in the NBF group than in the BF group (). However, CD4 cell count was determined at enrollment. Other variables that we assessed at each visit, such as HIV disease stage, were not significantly different between the BF and NBF women. Misclassification of self-reported BF history and some of the reported morbidity events (eg, reported general illnesses and hospital admissions) are potential weaknesses of this study.
How to protect the infant from complications related to NBF and keep the child HIV free is a dilemma in the settings where replacement feeding is not safe. BF provides several protective factors, meets nutritional requirements, and is critical for survival in developing countries, especially during the first months of life, as was shown in a large multi-country study [25
]. More options are now available to prevent mother-to-child transmission of HIV, ranging from extended infant prophylaxis to use of maternal antiretrovirals for prevention [14
]. More-potent drugs and regimens that start during pregnancy can lower further rates of transmission to levels seen in developed countries, as was recently demonstrated in a clinical trial conducted in Botswana [27
]. Making maximum use of current successful interventions to prevent HIV infection in children should provide the opportunity to prevent non–HIV infection–related morbidity and mortality among children.
Weaning is associated with acute and late serious adverse outcomes among HIV-exposed, uninfected children. Therefore, prolonged BF to ≥12 months should be encouraged, consistent with the recent WHO 2009 recommendations [13
]. With introduction of maternal antiretrovirals for both prophylaxis and treatment, BF duration could be extended. The impact of these changes on child health needs to be evaluated. In settings similar to Malawi, where the background morbidity and mortality are high [28
], inexpensive preventive strategies, such as TMP-SMX prophylaxis of HIV-exposed, uninfected children during and after weaning, should be considered.