Duration, pattern of breastfeeding and postnatal transmission of HIV: Pooled analysis of individual data from west and south African cohorts

Background: Both breastfeeding pattern and duration are associated with postnatal HIV acquisition; their relative contribution has not been reliably quantified. Methodology and Principal Findings: Pooled data from 2 cohorts: in urban West Africa where breastfeeding cessation at 4 months was recommended but exclusive breastfeeding was rare (Ditrame Plus, DP); in rural South Africa where high rates of exclusive breastfeeding were achieved, but with longer duration (Vertical Transmission Study, VTS). 18-months HIV postnatal transmission (PT) was estimated by Kaplan-Meier in infants who were HIV negative, and assumed uninfected, at age >1 month. Censoring with (to assess impact of mode of breastfeeding) and without (to assess effect of breastfeeding duration) breastfeeding cessation considered as a competing event. Of 1195 breastfed infants, not HIV-infected perinatally, 38% DP and 83% VTS children were still breastfed at age 6 months. By age 3 months, 66% of VTS children were exclusively breastfed since birth and 55% of DP infants predominantly breastfed (breastmilk+water-based drinks). 18-month PT risk (95%CI) in VTS was double that in DP: 9% (7-11) and 5% (3-8), respectively (p = 0.03). However, once duration of breastfeeding was allowed for in a competing risk analysis assuming that all children would have been breastfed for 18-month, the estimated PT risk was 16% (8-28) in DP and 14% (10-18) in VTS (p = 0.32). 18-months PT risk was 3.9% (2.3-6.5) among infants breastfed for less than 6 months, and 8.7% (6.8-11.0) among children breastfed for more than 6 months; crude hazard ratio (HR): 2.1 (1.2-3.7), p = 0.02; adjusted HR 1.8 (0.9-3.4), p = 0.06. In individual analyses of PT rates for specific breastfeeding durations, risks among children exclusively breastfed were very similar to those in children predominantly breastfed for the same period. Children exposed to solid foods during the first 2 months of life were 2.9 (1.1-8.0) times more likely to be infected postnatally than children never exposed to solids this early (adjusted competing risk analysis, p = 0.04). Conclusions: Breastfeeding duration is a major determinant of postnatal HIV transmission. The PT risk did not differ between exclusively and predominantly breastfed children; the negative effect of mixed breastfeeding with solids on PT were confirmed. © 2009 Becquet et al. The Ditrame Plus study was granted ethics permission in Cote d’Ivoire from the ethics committee of the National AIDS Control Programme, and in France from the institutional review board of the ANRS (Agence Nationale de Recherche sur le Sida et les hepatites virales). The Vertical Transmission Study was approved by the Biomedical Research Ethics Committee of the University of KwaZulu-Natal, South Africa. All women included in the Ditrame Plus study and in the Vertical Transmission Study provided written informed consent. We pooled together the data from two cohort studies, of which inclusion procedures and research design have been described in detail elsewhere: the Ditrame Plus study in Côte d’Ivoire [11], [12] and the Vertical Transmission Study (VTS) in South Africa [5], [13]. This pooled analysis was planned from the inception of both studies, which contributed to avoid large differences between studies, with regard to data collection and analysis [14]. This approach allowed for standardized study design and standardized definitions of exposure (see the subsequent “Collection of data on infant feeding practice” paragraph) and confounder variables in both individual studies. From March 2001 to July 2003, pregnant, HIV-infected, women aged ≥18 years from selected community-run health facilities were enrolled in the Ditrame Plus study in Abidjan, Côte d’Ivoire [11], [12]. All enrolled women received peri-partum antiretroviral zidovudine with or without lamivudine, and single-dose nevirapine to prevent mother-to-child transmission of HIV. Single-dose nevirapine was offered to all infants. Antenatally, two feeding options were discussed: complete avoidance of breastfeeding or exclusive breastfeeding with early cessation from the fourth month. Maternal infant feeding choice was supported, and replacement feeding from birth, or from breastfeeding cessation until 9 months of age, was provided free of charge. Mother-infant pairs were offered clinical, nutritional and psychosocial follow-up in study clinics from birth to two years. From August 2001 to August 2004, pregnant, HIV-infected, woman aged ≥16 years within selected clinics in rural, semi-urban and urban KwaZulu-Natal, South Africa were enrolled in the VTS [5], [13]. Single-dose nevirapine was offered to all women and infants. Women were counselled antenatally about infant feeding options in line with WHO guidelines [15]. Emphasis was placed on the promotion of exclusive breastfeeding for the first six months for women choosing breastfeeding. From 2002, a six months’ supply of infant formula was offered free through the KwaZulu-Natal prevention of mother-to-child transmission programme; women could choose to access this supply from birth or any time in the first 12 months after delivery. Breastfeeding mothers were visited at home by counsellors who supported them in their infant feeding choice every two weeks until the infant was aged 6 months. Clinical follow-up of infants from birth to two years took place in study clinics. Similar questionnaires were used to collect infant feeding practices in both studies [10], [16], [17]. As recommended [18]–[20], infant feeding practices were recorded as 24-hour and seven-day recall histories detailing all feeds, liquids and non-human milks given to infants for every day of the preceding week. In both studies, this information was recorded by trained field workers who were not involved in infant feeding counselling. Infant feeding practices were recorded weekly for the first 9 months, and then quarterly until age 24 months in the VTS (41 visits); and weekly until 6 weeks of age, monthly until 9 months of age, and every 3 months until the child’s second birthday in the Ditrame Plus Study (18 visits). For this analysis, infant feeding practices in the VTS were obtained only from visits shared with the Ditrame Plus schedule, taking the VTS visit closest to the age at which a Ditrame Plus monthly visit would have been made. We assessed the appropriateness of this in a sensitivity analysis for the VTS only which showed the same feeding behaviour despite the reduced number of visits used (data not shown). World Health Organization infant feeding definitions were used [21]–[23]. A child who was breastfed at a given age while having never received any other drink, food or non-human milk was exclusively breastfed. Breastfed children having been given water or water-based drinks only were predominantly breastfed from the date of introduction of these fluids. A child who was breastfed at a given age while also having received food-based fluids, solid foods or non-human milk was considered mixed fed from the date of introduction of these fluids or foods. In this paper we distinguish two types of mixed feeding: breastmilk and food-based fluids or solids; and breastmilk and non-human milk. In the Ditrame Plus Study, venous blood samples were collected within 72 hours of delivery, at 4 to 6 weeks of age, at 2, 3, 6 and 9 months of age, and then every 3 months until 18 months of age. An additional sample was obtained 2 months after breastfeeding cessation. HIV status was established by quantitative HIV RNA assay (Versant bDNA HIV RNA kit version 3.0, Bayer diagnostics, Emeryville, USA, and TaqMan HIV-1 RNA real-time PCR from 2003) [24], [25]. In the VTS, dried blood spot samples were obtained on filter paper within 72 hours of delivery, at 4 to 6 weeks of age, monthly until 9 months of age, and then every 3 months until 18 months of age. HIV status was established by quantitative HIV RNA assay (Nuclisens HIV-1 QT, Organon Teknika, Boxtel, Netherlands, and Nuclisens EasyQ HIV-1, Biomerieux, Boxtel, Netherlands) [26]. In both studies, paediatric HIV infection was defined as a positive plasma HIV-1 RNA PCR at any age. Children with a negative RNA PCR from a sample obtained at age ≥30 days who later became infected were considered to be HIV-infected postnatally [27], [28]. This analysis was conducted among live born children of HIV-infected mothers. We excluded infants not tested for HIV infection, with unknown timing of infection or infected during the peri-partum period, and with unknown or imprecise infant feeding practices (at least one recall history missing in the first 2 weeks of life) or formula-fed from birth. The Turnbull’s extension of the Kaplan-Meier procedure to interval-censored data is recommended to assess the postnatal risk of HIV transmission when the length of intervals between HIV tests is long. However a standard Kaplan-Meier approach to the analysis has been shown to be sufficient if most intervals are short (less than 3 months) [28], which was the case in our two studies (data not shown). We thus assessed cumulative HIV postnatal transmission in the first 18 months of life by Kaplan-Meier analysis; association with maternal and infant variables was quantified in a Cox regression analysis [29], [30]. Breastfeeding pattern and duration were considered simultaneously in Cox regression analyses. All multivariable analyses included a dummy variable for study (VTS vs. Ditrame Plus) to account for differing baseline risks between the two cohorts. Postnatal HIV transmission (beyond 4 weeks of age) was assessed in infants with a negative RNA PCR from a sample obtained at age ≥30 days. The time of acquisition of infection was assumed to be mid-way between the date of the last negative and the first positive HIV RNA test [27]. Survival analysis was conducted to compare HIV postnatal transmission between the two studies. We used two different approaches for censoring to investigate separately the effects of breastfeeding pattern and duration on the postnatal risk of HIV transmission. First, infants were censored on the last day they had been seen in study clinics or on the day they died; this approach allowed us to investigate the impact of breastfeeding duration on the postnatal risk of HIV transmission (as censoring did not depend on breastfeeding duration). Second, infants were censored on the date of breastfeeding cessation, unless they were still being breastfed at the end of the study or if the date of the last negative test was before the date of the breastfeeding cessation plus 30 days, in the latter two cases the infants were censored on the date of the last available negative HIV test. In this second approach, breastfeeding cessation was considered a competing event which allowed the comparison of the two cohorts with regard to the probability of acquisition of HIV postnatally at a given age, which assumes that all children of both studies had been breastfed until that age. This approach allowed us to study the impact of the breastfeeding pattern on the risk of HIV transmission, once the breastfeeding duration had been fully taken into account. Finally, we calculated smooth estimates of the hazard function taking into account the instantaneous age-specific incidence of postnatal transmission, using the penalized likelihood approach and proportional hazard regression model for interval-censored and left-truncated data (PHMPL) [31]. This analysis considered the competing risk from breastfeeding cessation, and the estimates obtained were expressed per 100 child-years of breastfeeding. Upper and lower confidence bands of the estimate of age-specific incidence were estimated using a Bayesian technique [31]. Statistical analyses were carried out using SAS software (version 9.1; SAS Institute, http://www.sas.com) and the PHMPL computer program developed by the INSERM Research Centre in Epidemiology and Biostatistics of the Bordeaux University (version 1.2; http://www.isped.u-bordeaux2.fr/recherche/biostats/Telechargement/PHMPL/us-Biostats-PHMPL.htm).