Why is multiple micronutrient powder ineffective at reducing anaemia among 12-24 month olds in Colombia? Evidence from a randomised controlled trial

Abstract

In Colombia's bottom socio-economic strata, 46.6% of children under two are anaemic. A prevalence of above 20% falls within the WHO guidelines for daily supplementation with multiple micronutrient powder (MNP). To evaluate the effect of daily MNP supplementation on anaemia amongst Colombian children aged 12-24 months we ran a cluster RCT (n=1440). In previous work, we found the intervention had no impact on haemoglobin or anaemia in this population. In this current paper, we investigate this null result and find it cannot be explained by an underpowered study design, inaccurate measurements, low adoption of and compliance with the intervention, or crowding out through dietary substitution. We conclude that our intervention was ineffective at reducing rates of childhood anaemia because MNP itself was inefficacious in our population, rather than poor implementation of or adherence to the planned intervention. Further analysis of our data and secondary data suggests that the evolution with age of childhood anaemia in Colombia, and its causes, appear different from those in settings where MNP has been effective. Firstly, rates of anaemia peak at much earlier ages and then fall rapidly. Secondly, anaemia that remains after the first year of life is relatively, and increasingly as children get older, unrelated to iron deficiency. We suggest that factors during gestation, birth, breastfeeding and early weaning may be important in explaining very high rates of anaemia in early infancy. However, the adverse effects of these factors appear to be largely mitigated by the introduction of solid foods that often include meat. This renders population wide MNP supplementation, provided after a diet of solid foods has become established, an ineffective instrument with which to target Colombia's childhood anaemia problem.

Keywords: Anaemia; Child; Colombia; Haemoglobin; Iron-deficiency; Micronutrients; Multiple micronutrient powder; Nutrition.

Fig. 1

Meta analysis of the effect of MNP interventions on mean haemoglobin. Random effects meta-analysis performed using STATA’s metan command. Age(months)=age in months at the start of intervention. WMD=Weighted Mean Difference. Hb(BL)=baseline haemoglobin (g/L). Hb(FU-control)=haemoglobin (g/L) of the control group at follow-up. Bottom row excludes our study. Studies weighted using effective sample size accounting for design effects in cluster randomised controlled trials. For studies that do not report the intra-cluster correlation coefficient (ICC) we use 0.07 (mid point between the two studies that report an ICC: Attanasio et al. (2014) and Jack et al. (2012)) to calculate the effective sample size. Details of all studies provided in Table A1.

Fig. 2

Coefficient plot of estimated effects of MNP on haemoglobin for subgroups of children with baseline haemoglobin below successively lower thresholds. For each baseline haemoglobin cut-off on the x-axis the plot shows the estimated effect of MNP on follow-up haemoglobin for the subgroup of children whose baseline haemoglobin was below that cut-off. Points represent point estimates and bars represent the 95% confidence interval around each estimate. All estimated effects adjusted for sex, tester, region, second order polynomials in age and altitude and baseline haemoglobin using linear regression. Confidence intervals adjusted for clustering at the town level.

Fig. 3

Prevalence of childhood anaemia by age across different regions. Colombian data from ENSIN 2010 for sub-sample SISBEN level one households. USA data from PEDNSS. Continent level data from all DHS surveys which collected childhood anaemia rates. Most recent data used when more than one survey available. Country data weighted by World Bank population estimates of children under four to estimate the combined prevalence, grouped at the continent level. Countries included in continent level averages are listed in the online appendix (Table A5).