The World Health Organization (WHO) has since 2001 recommended exclusive breastfeeding (EBF) for the first six months of life 1 . It is estimated that universal coverage with general nutritional interventions including exclusive breastfeeding promotion could prevent 8% of child deaths under the age of 36 months and 10-15% of stunting 2 . The systematic reviews by Kramer et al. reported that EBF for 6 months compared to EBF for 3 to 4 months resulted in lower diarrheal morbidity, prolonged lactational amenorrhoea and no clear infant growth deficit among infants in either low- or high income countries 3 4 . However, few African studies were included in that review and the authors stated that “larger sample sizes would be required to rule out modest increases in the risk of undernutrition” with longer duration than 3–4 months of EBF 3 5 .

A recent systematic review reported that peer support for breastfeeding decreased the risk of non-exclusive breastfeeding at last study-follow-up by 37% in low- and middle income countries compared to only 10% in high income countries 4 . Even if breastfeeding promotion can substantially increase the proportion of infants that are breastfed 6 7 , its effect on growth is less clear. It is assumed that breastfed infants are healthier than non-breastfed infants 2 8 9 . This is expected to result from reduced incidence and severity of communicable diseases, including diarrhoea 6 8 and improved feeding during illness 6 9 10 . Other expected benefits are lower risk of autoimmune diseases 11 and potential long term beneficial effects including on cognition 12 13 14 15 . In addition, formula fed infants may receive excess energy, which could lead to overweight in childhood, and potentially later in life 10 16 17 .

This cluster-randomised trial estimated the effect of community-based promotion of EBF by peer counsellors on the prevalence of EBF and diarrhea 18 . In order to address whether EBF promotion could impact infant growth, this paper reports on growth patterns including ponderal growth (expressed as weight-for-length z-scores) and linear growth (length-for-age z-scores) as well as weight-for-age z-scores up to 6 months of age in children participating in the PROMISE EBF trial in Burkina Faso, Uganda and South Africa. Our research question was whether there was any significant difference in growth patterns between the intervention and control clusters by 24 weeks of age.

The effect of peer counselling on EBF and diarrhoea prevalence and the methods used in the PROMISE EBF trial are described elsewhere 18 . Briefly, the study was conducted in three countries: Burkina Faso, Uganda and South Africa. In Burkina Faso, the study was conducted in Banfora, a south-western rural area dominated by subsistence farming. In Uganda, the sites comprised rural Bungokho in Mbale District where both subsistence farming and petty trading are common, and urban Mbale Municipality characterised by informal settlements and small industries. There were three geographically separate sites in South Africa: 1) Paarl, a commercial farming area in the Western Cape Province, 2) peri-urban Umlazi, and 3) rural Rietvlei in KwaZulu-Natal. Infant mortality rates (IMRs) at the time of the study were 92/1,000 in Burkina Faso and 85/1,000 in Uganda. In South Africa, the IMRs were 40/1,000 in Paarl, 60/1,000 in Umlazi and 99/1,000 in Rietvlei 19 .

Within each country and site, clusters were randomised 1:1 to either the intervention (provision of EBF peer-counselling) or the control arm (where EBF was not promoted by our research team) 18 . This was also the case in South Africa, and in addition a separate team of peer supporters supported the families to obtain birth certificates and social welfare grants in the control clusters 20 . This was believed not to interfere with breastfeeding behaviour. Clusters were selected geographically to reduce contamination of the intervention.

The intervention consisted of EBF counselling by peer-counsellors that were from the local communities and trained in a one week course by the national research teams with a curriculum adapted from the WHO courses ‘Breastfeeding Counselling: a Training Course’ 21 and ‘HIV & Infant Feeding Counselling: a Training Course’ 22 . All mothers were offered at least five home visits, the first occurred in the third trimester of pregnancy. Qualitative descriptions of the intervention are given elsewhere 23 24 25 .

Sample size calculation was done for EBF and diarrhea prevalence 18 . Those intending to breastfeed and planning to stay in the selected cluster in the forthcoming year were included, further details on recruitment, inclusion and exclusion are given in the Additional file 1. The analysis included 2,579 singleton live children. The trial profile is given in Figure 1.

Data were collected between 2006 and 2008 using a recruitment interview late in pregnancy, and further interviews targeted at weeks 3, 6, 12 and 24 after birth. Structured questionnaires were developed and adapted from the literature 26 27 and informed by prior work in the participating countries 28 29 30 . The first recruitment interview focused mainly on socio-demographic and socio-economic characteristics. The follow-up contact points captured mother-reported feeding practices and infant illness, and the children’s length and weight were recorded. Interviews were regarded as ‘timely’ if they were done within the following time periods: 1.5–4.5 (3); 4.5–9 (6); 9–18 (12); and 18–28 (24) weeks. Timely measurements provided data for the undernutrition prevalence estimates. The trial profile reflects number of ‘timely’ out of ‘total’ interviews per scheduled visit (Figure 1). The trial profile also reflects the number of anthropometric measurements, all measurements went into the longitudinal modelling of the anthropometric data.

Length was measured to the nearest 0.1 cm using ‘Shorr Height-Length Measuring Board’ (Maryland, USA) provided as ‘Baby/infant/adult Length-height measuring system SET 2’ from UNICEF in Uganda, and ‘Seca 210 mobile measuring mats for babies and toddlers,’ with 0.5 cm precision, in Burkina Faso and TALC’s roller mat infantometer (Oxford, UK) in South Africa. Infant weight was recorded to the nearest 0.1 kg using the ‘Infant scale spring type, 25 kg, 100 g’ from UNICEF in Uganda, the ‘SECA 872 scale with mother-infant children’s’ function’ in Burkina Faso and ‘Mascott electronic bucket’ in South Africa. The measurement procedures followed standard WHO guidelines 31 . All data collectors underwent training in content, questionnaire techniques and measurements and were kept uninformed about cluster allocation. For the weight and length measurements, reproducibility and validity exercises were conducted. Re-training and evaluation were done at least semi-annually during the data collection period. The acceptable technical error of measurement (TEM) for a data collector 32 was a value less than two times that of the data collector supervisor, a clinical specialist.

This paper presents growth patterns including ponderal and linear growth and weight information of children up to six months of age who participated in a community-based trial assessing promotion of EBF by peer counsellors in Burkina Faso, Uganda and South Africa. Although the reported EBF prevalence doubled at 12 weeks in the intervention arms in all three countries, the absolute increase was large in Burkina Faso and Uganda (44 and 38 percentage points, respectively) and small in South Africa (4 percentage points) 18 . The child growth patterns varied in the three countries, with South Africa having the highest z-scores on average and Burkina Faso having the lowest. Ponderal growth tended to be slightly lower in Burkina Faso and Uganda in the intervention arms compared to the control arm even if the absolute weight differences were quite small at 24 weeks, around 100 g and statistically not significant in Burkina Faso, and around 200 g in Uganda at six months. Wasting was also more common at 24 weeks of age in Uganda. No significant differences were found for LAZ-scores between children in the intervention and control arms.

Compared to individual randomisation, cluster randomisation is more vulnerable to suboptimal randomisation as fewer units are randomised, and there is a chance that particular characteristics may be clustered 36 . However, the intra-cluster coefficients for the main outcomes were small 18 . Residual confounding from factors which we were not able to assess is still possible; for instance, as many of the births took place at home, gestational age and birth weight were available only for a small proportion of the mother-infant pairs in Burkina Faso and Uganda. Some would argue that our anthropometric measurements at around 3 weeks could act as a proxy for the corresponding baseline characteristics. However, there were only small differences in WLZ at 3 weeks of age and adjusting for 3 week weight (data not shown) did not substantially alter our effect estimates. It is known that societal, maternal and individual factors are related to growth outcomes 35 37 38 , and some of these may have been unmeasured and unevenly distributed between the study arms. However, adjusting for socio-economic status, which is likely to capture some of these characteristics, only minimally altered our effect measures. Missing data were most common at 3 weeks because many mothers left their homes and stayed with their relatives for some time after giving birth. An inverse-probability weighted method yielded similar results compared to an available-subject analysis (data not shown) indicating no noteworthy bias from missing data 39 .

There were major country differences with regard to the effect of the intervention on EBF prevalence 18 , and as reported in this paper, socio-economy, maternal education and BMI as well as in infant growth patterns. There were also country differences with respect to perinatal mortality 40 41 42 . Thus, we find it most appropriate to present the results by country although pooling the data would have increased our statistical precision. As the absolute difference in EBF prevalence between the arms in South Africa was very small, it is difficult to attribute any differences in growth patterns to the EBF promotion. The country specific contextual challenges explaining this low uptake of EBF has been described 43 as well as poor breastfeeding practices 44 . The peer support for families to obtain a social welfare grant provided in the control clusters is also unlikely to have mitigated child growth.

The infants in the Multicentre Growth Reference Study (MGRS) study 45 , which yielded the reference against which our infants’ growth was assessed, had non-smoking mothers from middle class or ‘affluent’ environments supportive of healthy growth 46 . In that study, 75% were exclusively or predominantly breastfed for 4 months and nearly 70% breastfed for a year. In our study, children from both Burkina Faso and Uganda were at 12 weeks of age exclusively breastfed to the same or even to a higher extent. The children in the PROMISE EBF trial were believed to represent the general population from their respective study areas and were not purposely selected from well-to-do families. In both arms of the PROMISE EBF trial in Burkina Faso and Uganda, we observed a gradual and substantial shift of the distributions towards poorer linear growth with increasing age, with a mean LAZ between −0.6 and −0.9 at 24 weeks. This growth pattern is described also in other studies in sub-Saharan Africa 37 .

In Burkina Faso and Uganda, the prevalence of wasting was slightly higher in the intervention arms as compared to the control arms at 12 and 24 weeks. This finding informs the debate launched by Kramer and colleagues who also found an average weight reduction associated with EBF, and could not rule out an increased risk of undernutrition 3 5 . This could indicate that our intervention was inappropriate for the most vulnerable children. Even if the difference in mean WLZ of the children in the intervention and the control arms in Burkina Faso and Uganda was similar at 24 weeks, the distribution of the WLZ of the children in the intervention arm in Burkina Faso was skewed towards lower values, away from the WHO growth standard mean, while in Uganda, the mean WLZ among children in the intervention arm was closer to the WHO WLZ mean. A shift towards lower WLZ might benefit populations in which obesity is common 16 , but in Uganda, where most communities are challenged by widespread undernutrition, the long term health consequences of a possible impairment in ponderal growth could be a concern.

The one week training course equipped the peer counsellors with basic information on promoting and supporting EBF, thus increasing EBF prevalences substantially in Burkina Faso and Uganda 24 . However, the peer counsellors had relatively low educational level and did not have other training in health care. Further, they often operated in environments with limited infrastructure with respect to water and sanitation, and where the public health system was not optimal, accessible and equitable, contributing to wasting, stunting and underweight 47 . Further research is needed to address the role and qualifications of peer counsellors for EBF to provide them with support to deliver safe interventions 30 .

Our research group has described, particularly in Uganda, how poverty and food insecurity is an important challenge to proper child feeding 48 49 . This is also a problem in Burkina Faso where the anthropometric status of children seems to be even worse 50 . It is also known that formula feeding is mostly unaffordable, unacceptable and unfeasible in both Burkina Faso and Uganda 49 , so access to industry formula cannot explain any of our findings. Neither can increased diarrhoea morbidity 18 . Even if the intervention has been described as acceptable by mothers in Uganda 51 , it is not fully understood how the intervention altered feeding behaviour. However, a recent quantitative description from the Promise EBF trial on changes in feeding categories at the different time points 52 informs us that there are shifts in all directions with time between the different feeding categories. It is important that future studies address how an infant feeding intervention might change both maternal and infant diets and behaviour.

Breastfeeding of infants up to 6 months of age was nearly universal in both Burkina Faso and Uganda 53 . This intervention might mainly have shifted infants from being predominantly or partially breastfed to being exclusively breastfed 52 . However, it is important to assess how the observed changes have impacted on morbidity and mortality. Some studies highlight that a shift from predominant to exclusive breastfeeding up to six months does not add any health benefits 9 . A trial in Guinea-Bissau, despite challenges with high losses to follow-up, saw higher EBF prevalence in the intervention arm, but did not find a beneficial impact on diarrhoea prevalence, mortality or infant weight 54 . In fact, in the subset (40%) of children that were weighed close to 6 months of age, the median weight in the intervention group was slightly lower than in the control group. Likewise, our PROMISE EBF trial found no effect on diarrhoea morbidity 18 . Assessment of mortality did not indicate clear differences but were not powered for comparing the trial arms 40 41 . Although our trial protocol did not calculate sample size for child growth we think the sample size was sufficient for this intention-to-treat analysis on growth patterns. This is supported by a post hoc power calculation indicating high power, the narrow confidence intervals (Additional file 1) and high follow-up rates (Figure 1).

Follow-up studies are needed to assess long-term growth and health patterns for these children in order to balance potential negative and positive effects of EBF promotion is these settings. This will inform WHO’s global nutrition target towards 2025 aiming at reducing malnutrition 55 .

There were small differences in growth patterns between the study arms in the three countries. In both Burkina Faso and Uganda, children in the intervention clusters had slightly lower ponderal growth at 24 weeks of age. Beneficial health effects or absence of negative impact on morbidity, mortality and infant growth reported in some EBF promotion trials in Asia have yet to be demonstrated in Africa. There is a need to better understand 1) the reasons for this discrepancy, 2) how to optimize infant feeding support in Sub-Saharan Africa, where predominant breastfeeding and undernutrition is common, and 3) how to ensure that breastfeeding promotion can effectively be combined with appropriate referral of vulnerable children for nutritional support and disease management.

CI: Confidence intervals; EBF: Exclusive breastfeeding; LAZ: Length-for-age z-scores; LSM: Least Squares Means; MGRS: Multicentre Growth Reference Study; WAZ: Weight-for-age z-scores; WHO: World Health Organization; WLZ: Weight-for-length z-scores.

The authors declare that they have no competing interests.

Among the authors, IMSE, DJ, AHD, TD, JN, DS, NM, JKT, ECE, PVdP, CK, HS and TT designed the study and the intervention. IMSE, DJ, LTF, VN, TT and HS planned and wrote the paper. IMSE and LTF handled, cleaned and analysed the data. SS and CL were study statisticians. DS contributed to manuscript design and content. AHD, VN and TD had particular responsibility for study implementation and data quality and management in Burkina Faso, Uganda and South Africa, respectively. HW contributed to data management and VR contributed to analytical content. ECE, NM, JKT, PvP, CK and DJ were country principal investigators and planned the study design, administered implementation and worked on analytic content. TT was the central PI. All authors read and contributed towards the final manuscript.