Effects of early feeding on growth velocity and overweight/obesity in a cohort of HIV unexposed South African infants and children

Background: South Africa has the highest prevalence of overweight/obesity in Sub-Saharan Africa. Assessing the effect of modifiable factors such as early infant feeding on growth velocity and overweight/obesity is therefore important. This paper aimed to assess the effect of infant feeding in the transitional period (12 weeks) on 12-24 week growth velocity amongst HIV unexposed children using WHO growth velocity standards and on the age and sex adjusted body mass index (BMI) Z-score distribution at 2 years. Methods: Data were from 3 sites in South Africa participating in the PROMISE-EBF trial. We calculated growth velocity Z-scores using the WHO growth standards and assessed feeding practices using 24-hour and 7-day recall data. We used quantile regression to study the associations between 12 week infant feeding and 12-24 week weight velocity (WVZ) with BMI-for-age Z-score at 2 years. We included the internal sample quantiles (70th and 90th centiles) that approximated the reference cut-offs of +2 (corresponding to overweight) and +3 (corresponding to obesity) of the 2 year BMI-for-age Z-scores. Results: At the 2-year visit, 641 children were analysed (median age 22 months, IQR: 17-26 months). Thirty percent were overweight while 8.7% were obese. Children not breastfed at 12 weeks had higher 12-24 week mean WVZ and were more overweight and obese at 2 years. In the quantile regression, children not breastfed at 12 weeks had a 0.37 (95% CI 0.07, 0.66) increment in BMI-for-age Z-score at the 50th sample quantile compared to breast-fed children. This difference in BMI-for-age Z-score increased to 0.46 (95% CI 0.18, 0.74) at the 70th quantile and 0.68 (95% CI 0.41, 0.94) at the 90th quantile. The 12-24 week WVZ had a uniform independent effect across the same quantiles. Conclusions: This study demonstrates that the first 6 months of life is a critical period in the development of childhood overweight and obesity. Interventions targeted at modifiable factors such as early infant feeding practices may reduce the risks of rapid weight gain and subsequent childhood overweight/obesity. The present paper includes data from the three South African sites of the PROMISE-EBF behavioural-intervention trial that sought to improve EBF rates through peer counselling, conducted between 2006 and January 2008: Paarl (mixed peri-urban/rural area), Rietvlei (rural area) and Umlazi (peri-urban formal township). Trial methods have been described in detail elsewhere [28,29]. Briefly, pregnant women in their last trimester of pregnancy were screened for inclusion into the study. A total of 964 HIV negative and 184 HIV positive women and their singleton children were enrolled at the 3 week postnatal visit and followed up at 6, 12 and 24 weeks . Six hundred and fifty four HIV unexposed children (67.8% of original cohort) followed up again between March and September 2008 at a median age of 22 months (IQR: 9–34 months), which we refer to as the 2 year visit, were considered for this analysis due to the described negative effect of HIV infection on growth [27]. We compared baseline characteristics of participants that were followed-up at 2 years with those that were not (see Additional file 1) and observed no systematic differences, besides the proportion of male children, between the groups. This suggests that the sample that was followed-up is generally representative of the children in the whole cohort. A further 13 children were excluded because of extreme and implausible anthropometric values leading to a final sample of 641 children. Standardised questionnaires were used to collect interview data during pregnancy and postnatally at 3, 6, 12 and 24 weeks, and at the primary endpoint of 2 years of age. Maternal variables included: age, parity and education which were captured during recruitment; delivery mode and reported HIV status collected at the 3 week visit. The questionnaires also addressed infant feeding practices through 24-hour and 7-day recall of a list of 23 foods commonly consumed in the study sites. No food diaries were used. Data on child birth weight were extracted from perinatal records. Field staff measured child weight and recumbent length/height during the 3, 6, 12, and 24 weeks visits and at 2 years. Children were weighed to the nearest 0.1 kg on Masskot (SOS Series) electronic pan scales, which were calibrated weekly using a 2 kg weight, wearing minimum clothing and no shoes. Depending on the study site, recumbent length measurements were obtained to the nearest 0.1 cm using TALC roller meters (Oxford, UK) or Shorr Height-Length Measuring Board (Maryland, USA) while height was measured using a validated ustom-made stadiometer. All field workers were trained on anthropometric techniques. In order to improve validity and reduce inter and intra-observer bias, the anthropometry data collection was validated periodically. Child age was calculated using the date of birth from the Road to Health card and the date of the interview. Data were double-entered into a Microsoft Access database and analysed using Stata SE 12 [30] and IBM SPSS Statistics 21 [31]. The primary outcome measure was BMI-for-age Z-scores at 2 years; secondary outcomes were weight velocity Z-scores (WVZ) and length velocity Z-scores (LVZ). We calculated BMI-for-age Z-scores at the 12 week and 2 year visits, standardised for sex and actual age at the respective visit, using the WHO growth standards [32]. We considered children as “overweight” and “obese” if their BMI-for-age Z-scores were above +2 and +3 respectively as recommended by the World Health Organisation [33]. A macro based on the WHO-2009 growth velocity standards was used to compute the WVZ and LVZ. Velocities were calculated for a first period, namely from 3 or 6 to 12 weeks post-delivery, and for a second period, namely 12 to 24 weeks post-delivery. In cases where the 3 or 6 week weight was missing we used the birth weight for the calculation of velocity in the first period. The age intervals and child ages observed in the study did not always correspond exactly with those of the velocity standards. Thus the velocity Z-scores were calculated, as recommended by WHO, by identifying the best-fitting age interval for each child period observed and linearly extrapolating the observed increment in the child period to the duration of the best-fitting target interval [26]. Anthropometric measurement values and Z-scores were flagged for verification if any of the following criteria were met: a) decrease in length of more than 2 cm between two consecutive visits; b) WAZ <-6 or >5, WLZ <-5 or >5, LAZ <-6 or >6, WLZ >3 and LAZ <-3; c) extreme changes in LAZ between visits defined as LAZ at 3 weeks < 2 and LAZ at 24 weeks > 2.5, or LAZ at 24 weeks < 2 and LAZ at 36 weeks > 2.5; d) changes > 4 or <-4 Z-scores between 24 and 36 weeks and BMI-for-age Z-score ≥6. All the flagged anthropometric observations were assessed and values treated as missing if no plausible explanation was determined. We used a combination of 24-hour and 7-day infant feeding recall data at each follow-up visit to generate time specific food consumption indicator variables (for breast milk, water, sugar water, formula, cereals, fruits/vegetables, traditional herbs, prescribed and non-prescribed medicines) with 3 categories: yes, no and missing. For example if the caregiver said that she gave the child breast milk in the previous 24-hours or 7-days then we coded that child as having received breast milk. If the caregiver said “no” to both questions on breast milk, the child was then considered as one that did not receive breast milk. The response was coded as “missing” for breast milk if data were missing for both questions. Cross-tabulation of the 12 week breast milk and formula indicators revealed that all children had consumed at least one of the two foods. Based on exploratory analysis we combined two of the three combinations of these feeding indicators and this resulted in a binary ‘ever breastfed” variable with the following categories: yes (received breast milk with other solids and liquids which may include formula) and no (received formula and other liquids and solids except breast milk). The 12 week breastfeeding cessation variable was defined as no breastfeeding at the 12 week interview (based on 24-hour and 7-day recall) and no breastfeeding reported for the subsequent final 24-week interview. Only children who initiated breastfeeding by the 3 week visit were considered in this definition. Unlike the ordinary least squares (OLS) regression which only considers the conditional mean function, we used quantile regression which is a statistical technique that provides a more detailed analysis of the relationship between the dependent variable and its independent variables because it provides conditional regression coefficients for each quantile, [34,35]. We used univariate and multivariate simultaneous quantile regression to test whether 12 week infant feeding and 12–24 week growth velocity (adjusting for other variables) had increased effects over the upper tails of the conditional distribution of BMI-for-age Z-scores at 2 years. For this analysis we included the internal sample quantiles (70th and 90th centiles) that approximate the reference cut-offs of +2 and +3 Z-scores for BMI-for-age around 2 years. We also performed OLS regression modelling. The following variables were adjusted for in the multivariate models because of their epidemiologic or clinical importance: birth weight, maternal age, parity, maternal education, study arm and site. Although the child’s age and sex were taken into account in the BMI-for-age Z-score estimations, based on previous literature [36] we included an interaction term between the infant feeding and sex variables in initial regression models to test whether sex modifies the relationship between feeding and BMI-for-age Z-score. This interaction term was excluded from the final models as no effect measure modification was detected. Maternal age and parity were excluded from the final multivariate model because they were not significantly associated with 2-year BMI in the univariate analysis. The Breusch-Pagan / Cook-Weisberg test was used to check for heteroskedasticity and trends across the quantile regression percentiles were also tested. Continuous data are presented as mean ± SD or median (IQR) while categorical variables are presented as frequencies. We used the Student t-test to compare means and the Pearson chi-square test to examine associations in the cross-tabulations. Statistical tests were two-sided and performed at the 5% significance level. Kernel density functions were used to estimate the 2 year BMI-for-age Z-score distribution stratified by the 12 week overweight and breastfeeding while the two-sample Kolmogorov-Smirnov test was used to test for equality of the distribution functions. The PROMISE-EBF trial was approved by the Regional Committees for Medical and Health Research Ethics (REK VEST) in Norway (issue number 05/8197), University of the Western Cape (research registration number 0607/8) and the South African Medical Research Council (protocol ID: ECO7-001). Informed consent was obtained from all participants.