Formula-Feeding of HIV-Exposed Uninfected African Children Is Associated with Faster Growth in Length during the First 6 Months of Life in the Kesho Bora Study

Background: Early feeding patterns may affect the growth of HIV-exposed children and thus their subsequent health and cognition.Objective: We assessed the association of infant feeding (IF) mode with length-for-age z score (LAZ) and stunting from age 2 d to 18 mo in HIV-exposed African children within a controlled randomized trial, which evaluated triple antiretrovirals initiated during pregnancy and continued for 6 mo postpartum to prevent HIV transmission.Methods: HIV-infected pregnant women with CD4+ counts of 200-500 cells/mm3 from Burkina Faso, Kenya, and South Africa were advised to exclusively breastfeed for up to 6 mo or to formula-feed from birth. Factors associated with LAZ were investigated in all uninfected children by using mixed-effects linear models; those associated with stunting (LAZ <-2) at 6 or 12 mo were assessed in multiple logistic regression after exclusion of children stunted at age 2 d. Independent variables were IF mode: formula feeding (FF), exclusive breastfeeding (EBF) <3 mo, or EBF ≥3 mo (reference); sex; trial arm; maternal characteristics; and site.Results: Among 728 children, FF was associated with a greater increase in LAZ from 2 d to 6 mo (+0.07 z score/mo, P < 0.001). Between 6 and 18 mo, FF and EBF <3 mo were both associated with greater mean LAZ than was EBF ≥3 mo (+0.52 z scores and +0.43 z scores, respectively, P < 0.001). Among children not stunted at 2 d, FF was independently associated with a reduced risk of stunting at 6 mo (OR: 0.24; 95% CI: 0.07, 0.81; P = 0.021), whereas EBF <3 mo was not (OR: 0.49; 95% CI: 0.22, 1.10; P = 0.09).Conclusions: In this observational study of HIV-exposed uninfected infants, growth in length in the first 6 mo of life was faster in formula-fed infants than in exclusively breastfed infants. The plausibility of residual confounding and reverse causality is discussed. This trial was registered at www.controlled-trials.com as ISRCTN71468401. The Kesho Bora RCT was designed to test the efficacy of an intervention to reduce mother-to-child transmission (MTCT) of HIV and to increase infant and maternal survival (13). The intervention group received a combination of 3 antiretroviral drugs from 28 to 36 wk of gestation until 6 mo postpartum (or breastfeeding cessation, if earlier). The control group received a standard short-course MTCT-preventive antiretroviral regimen (13). Women were recommended to either breastfeed exclusively up to 6 mo postpartum or formula-feed from delivery, in accordance with WHO recommendations and national guidelines at the time (14, 15). All live-born singletons or first-born twins of enrolled women were included in the present analysis if they had ≥1 length measurement, were HIV-uninfected at the end of follow-up, and had no missing data for potentially confounding factors (i.e., study site, RCT arm, sex, wealth indicator, maternal education, height, and BMI). The study was conducted in Bobo-Dioulasso, Burkina Faso; Nairobi and Mombasa, Kenya; and Durban and a rural area of KwaZulu-Natal, South Africa. Enrollment started in January 2005 in the 3 former sites and in February and September 2007 in Durban and in Somkhele, rural KwaZulu-Natal, respectively. Enrollment was completed in August 2008 and follow-up in December 2010, respectively (13). Ethical clearance was given by the ethical and regulatory committees in Burkina Faso, Kenya, and South Africa, and at the WHO and CDC. The study was conducted in accordance with the Helsinki Declaration of 1975, as revised in 1983, and all women provided written, informed consent. Pregnant, HIV-infected women at <32 wk of gestation who did not participate in any other study and who had a WHO clinical stage 1, 2, or 3 disease and a CD4+ count of 200–500 cells/mm3 were eligible for enrollment. The triple-antiretroviral group received a combination of 3 antiretrovirals (zidovudine, lamivudine, lopinavir/ritonavir) from 28 to 36 wk of pregnancy up to 6.5 mo postpartum (or breastfeeding cessation, if earlier). Those in the short-course arm received standard prophylaxis, i.e., zidovudine and a single dose of nevirapine. All infants received a dose of nevirapine, preferably within 72 h of birth. The study was not blinded. Infant feeding counseling was conducted antenatally and after delivery until complete cessation of breastfeeding, following WHO guidelines. Demonstrations of formula preparations were performed and free formula was provided for infants from birth until 6 mo of age, as appropriate. All medical care was provided free of charge. Maternal background data were collected at enrollment (i.e., during pregnancy). Gestational age estimation was based on self-reported last menstrual period or symphysis-fundal height. Birth weight was measured routinely in maternity wards by hospital staff to the nearest 100 g; it was unavailable for home deliveries. Length measurements were not taken at birth. Early neonatal recumbent length was measured, preferably within 48 h and at most 7 d after delivery. Subsequent length measurements were taken in study clinics at ages 4 and 8 wk and 4, 6, 9, 12, 15, and 18 mo. All measurements were read to the nearest millimeter by trained teams using infant measuring boards and standard procedures (16). Two readings were made for each measurement, and the mean was used in the analysis. If the difference between the 2 readings exceeded 5 mm, a third reading was made and the outlier was discarded. Weight was recorded to the nearest 10 g by using electronic infant scales. IF data were collected during postnatal visits at study clinics at ages 2, 4, 6, and 8 wk and 3, 4, 5, and 6 mo, including all items fed to the infant since the preceding visit (since birth for the 2-wk visit). Continued breastfeeding was recorded at all clinic visits, including at ages 9, 12, 15, and 18 mo. In case of breastfeeding cessation since the preceding visit, the precise date of cessation was recorded. Blood samples for diagnosis of child HIV infection status were taken at birth, ages 2 and 6 wk, and 3, 6, 9, 12, and 18 mo (17). The HIV infection status of all children was assessed by quantitative RNA PCR tests at ages 6 wk and 12 mo (18). Samples taken at other ages were used for HIV diagnosis in those who died or were lost to follow-up before age 12 mo. All children reaching age 18 mo had their HIV infection status confirmed through HIV serology. The original protocol included follow-up until age 24 mo, but financial constraints reduced this to ages 12–18 mo during the course of the RCT. As a result, scheduled child follow-up changed over time; during the last year of the trial, child follow-up in the South African sites (i.e., Somkhele and Durban) was at age 12 mo only. Length-for-age and weight-for-length were expressed in z scores relative to the WHO growth standards by using the Anthro software (19). Unreliable length values (i.e., LAZ >3 or LAZ <−6) were excluded. Stunting was defined as LAZ <−2 and wasting as weight-for-length z score (WLZ) <−2. IF was divided into 3 groups: never breastfed (i.e., exclusively formula-fed from birth), duration of exclusive breastfeeding (EBF) <3 mo, or EBF ≥3 mo (6, 7). Breastfed children who died or were lost to follow-up before 3 mo were classified as EBF <3 mo, as were those who were breastfed but not exclusively. Maternal education was divided into 3 groups (none, some primary, and some secondary), and maternal height was divided into 4 arbitrarily defined categories (Table 1). Maternal postpartum BMI was computed by using the weight measurement taken at 4 wk postpartum or, if missing, at 6 or 8 wk. BMI (in kg/m2) was divided into mild underweight, lower-normal range, upper-normal range, and overweight-obese (<20, 20–22.4, 22.5–25, and >25, respectively), because the prevalence of true underweight (i.e., BMI <18.5) was very low (20). Maternal CD4+ cell count at inclusion was divided into 2 groups, i.e., 200–349 and 350–500 cells/mm3. Selected household possessions (refrigerator, radio, television, cell phone) and characteristics of housing (access to gas or electricity for cooking, clean drinking water, electricity, flush toilets, and sharing of toilets with persons from outside the household) were used to compute a summary economic score by multiple correspondence analysis. All children had complete data for these variables. Site-specific tertiles were used in the analysis (6). Characteristics of women and children in the Kesho Bora RCT by IF modality1 Low birth weight was defined as birth weight <2500 g. Gestational age at delivery was defined as gestational age at enrollment plus the number of weeks of follow-up before delivery. Because of a high rate of unreliable values (e.g., ≥44 wk), this variable was not used in the analysis. Maternal and child characteristics were compared across IF modalities and study sites by using chi-square or Fisher’s exact tests. Adjustment for study site in the analysis of IF modalities was carried out by using Cochran-Mantel-Haenszel tests. Chi-square tests and Cochran-Mantel-Haenszel tests were used to compare children with a length measurement at age 18 mo and those who stopped follow-up between 12 and 18 mo. Factors potentially associated with the hazards of becoming stunted at 6 or 12 mo for children not stunted at age 2 d were assessed in multiple logistic regression analysis. The main explanatory variable was IF, and the analysis was adjusted for the RCT arm, sex, and study site and maternal height, BMI, and wealth index (all categorical). Children already stunted at age 2 d were excluded from this analysis because early neonatal stunting cannot be attributed to IF. The existence of effect modification by study site was tested through the introduction of an interaction term between site and IF in a separate logistic regression model. Mixed-effects linear models (MLMs) with an autoregressive covariance structure were used to examine the associations between IF modality and LAZ. These analyses included all children, irrespective of stunting status at age 2 d. Models included random intercepts for each participant. Linearity was determined graphically. Potential confounding factors were RCT arm, study site, and infant sex and maternal height, postpartum BMI, level of education, CD4+ cell count at inclusion, and wealth index. The age intervals of 2 d to 6 mo and 6–18 mo were considered separately per protocol. Again, effect modification by study site was tested by the introduction of an interaction term between study site and IF in separate models. For each independent variable, MLMs provide 2 coefficients, β0 and β1. The former estimates the difference between a given category and its reference category at the start of the interval, whereas the latter assesses whether this difference is constant over time through an interaction term with age. If significant, the latter indicates that the slope of the category under study differs from that of the reference category. The estimated difference of a category with the reference at time 1 thus equals (β0 + β1 × time 1). In sensitivity analyses of logistic regression and MLM, IF was used as a binary variable (never- compared with ever-breastfed children). Analyses were performed by using SAS (version 9.3) and Stata (version 14, Stata Corp) statistical software, and differences were considered significant at P < 0.05. All tests were 2-sided. Distributions of continuous variables are provided as means ± SDs, and proportions as P (95% CIs).