Maternal and child supplementation with lipid-based nutrient supplements, but not child supplementation alone, decreases self-reported household food insecurity in some settings

Background: It is unknown whether self-reported measures of household food insecurity change in response to foodbased nutrient supplementation. Objective: We assessed the impacts of providing lipid-based nutrient supplements (LNSs) to women during pregnancy and postpartum and/or to their children on self-reported household food insecurity in Malawi [DOSE and DYAD trial in Malawi (DYAD-M)], Ghana [DYAD trial in Ghana (DYAD-G)], and Bangladesh [Rang-Din Nutrition Study (RDNS) trial]. Methods: Longitudinal household food-insecurity data were collected during 3 individually randomized trials and 1 clusterrandomized trial testing the efficacy or effectiveness of LNSs (generally 118 kcal/d). Seasonally adjusted Household Food Insecurity Access Scale (HFIAS) scores were constructed for 1127 DOSE households, 732 DYAD-M households, 1109 DYAD-G households, and 3671 RDNS households. The impact of providing LNSs to women during pregnancy and the first 6 mo postpartum and/or to their children from 6 to 18-24 mo on seasonally adjusted HFIAS scores was assessed by using negative binomial models (DOSE, DYAD-M, and DYAD-G trials) and mixed-effect negative binomial models (RDNS trial). Results: In the DOSE and DYAD-G trials, seasonally adjusted HFIAS scores were not different between the LNS and non-LNS groups. In the DYAD-M trial, the average household food-insecurity scores were 14% lower (P = 0.01) in LNS households than in non-LNS households. In the RDNS trial, compared with non-LNS households, food-insecurity scores were 17% lower (P = 0.02) during pregnancy and the first 6 mo postpartum and 15% lower (P = 0.02) at 6-24 mo postpartum in LNS households. Conclusions: The daily provision of LNSs to mothers and their children throughout much of the ”first 1000 d” may improve household food security in some settings, which could be viewed as an additional benefit that may accrue in households should policy makers choose to invest in LNSs to promote child growth and development. The study designs for the 4 randomized trials have been described elsewhere in detail (12, 15, 18, 20) and are summarized in Table 1. The DOSE trial was designed to test the efficacy of various doses and formulations of LNSs for promoting child growth. Rolling enrollment of children was conducted from November 2009 to May 2011. At ∼6 mo of age, children were randomly assigned to 1 of 5 intervention groups or a delayed-intervention control group. Children in the intervention groups received daily LNSs for 12 mo in one of the following doses and formulations: 1) 10 g LNS containing milk powder, 2) 20 g LNS without milk powder, 3) 20 g LNS containing milk powder, 4) 40 g LNS without milk powder, or 5) 40 g LNS containing milk powder. The delayed-intervention control group received no supplementation during the 12-mo intervention period. All children in the trial, regardless of intervention group, received weekly morbidity surveillance and referral by study staff. Study designs1 A pair of randomized controlled trials known as the DYAD trials were conducted in Malawi (DYAD-M) and Ghana (DYAD-G) to test the efficacy of LNSs provided to women during pregnancy and the first 6 mo postpartum and to children from 6 to 18 mo of age on birth outcomes and child growth. Rolling enrollment of pregnant women who were at <20 wk gestation occurred between February 2011 and August 2012 in DYAD-M and between December 2009 and December 2011 in DYAD-G. In both trials, women were randomly assigned to 1 of 3 intervention groups. During pregnancy and the first 6 mo postpartum, women in one group received a daily dose of 20 g LNS product designed for maternal consumption, and from 6 to 18 mo of age, the children of women in that group also received a daily dose of 20 g LNS product designed for children. Another group received a daily multiple micronutrient capsule during pregnancy and the first 6 mo postpartum (no child supplementation). A third group received a daily iron-folic acid (IFA) capsule during pregnancy, which is a component of the standard of antenatal care in Ghana and Malawi, and a daily 200-mg Ca placebo for the first 6 mo postpartum (no child supplementation). All mothers and children participating in the DYAD-M and DYAD-G trials also received biweekly (mothers) or weekly (children) morbidity surveillance and referral by study staff. The RDNS effectiveness trial in Bangladesh was implemented in partnership with LAMB, a nongovernmental organization that provided community-based health services to women and children in the study area. RDNS was a cluster-randomized trial with 4 intervention groups, with clusters defined based on the LAMB community health workers’ 64 work areas. Rolling enrollment of pregnant women who were at ≤20 wk gestation occurred between October 2011 and August 2012. Women randomly assigned to the comprehensive LNS group received 20 g LNS/d through pregnancy and the first 6 mo postpartum, and their children received 20 g LNS/d from 6 to 24 mo of age. Another group of women received daily IFA capsules during pregnancy and the first 3 mo postpartum, and their children received LNSs from 6 to 24 mo. A third group also received IFA capsules during pregnancy and the first 3 mo postpartum, and their children received micronutrient powder from 6 to 24 mo. A final group served as the control and received IFA during pregnancy and the first 3 mo postpartum (no child supplementation). All supplements were delivered by LAMB community health workers. The combined groups in Table 1 are based on pooling the intervention groups into 2 groups/trial: one group that received LNS and another that did not. In this article, the primary analysis was conducted with the combined groups, and a secondary analysis with the original assigned intervention groups was conducted as a sensitivity analysis. All trials received ethical approval and were registered at clinicaltrials.gov. The DOSE ({"type":"clinical-trial","attrs":{"text":"NCT00945698","term_id":"NCT00945698"}}NCT00945698) and DYAD-M ({"type":"clinical-trial","attrs":{"text":"NCT01239693","term_id":"NCT01239693"}}NCT01239693) trials were approved by the Research and Ethics Committee of the University of Malawi College of Medicine and by the Ethics Committee of Pirkanmaa Hospital District, Finland. The DYAD-G trial ({"type":"clinical-trial","attrs":{"text":"NCT00970866","term_id":"NCT00970866"}}NCT00970866) was approved by the ethics committees of the University of California, Davis; the Ghana Health Service; and the University of Ghana Noguchi Memorial Institute for Medical Research. The RDNS trial ({"type":"clinical-trial","attrs":{"text":"NCT01715038","term_id":"NCT01715038"}}NCT01715038) was approved by the ethics committees of the University of California, Davis; the International Center for Diarrheal Disease Research, Bangladesh; and LAMB. The nutrient contents of all capsules and LNS products used in the 4 trials are available in Supplemental Tables 1–3. Both the DOSE and DYAD-M trials were conducted in the Mangochi District in southern Malawi. DOSE participants were drawn from the semi-urban catchment area of the Mangochi district hospital and the rural catchment area of the Namwera health center. Women were recruited into the DYAD-M trial from the Malindi hospital (semi-urban) and Lungwena health center (rural) in addition to the 2 DOSE recruitment sites. The majority of the population in the Malawi study catchment areas were subsistence farmers and fishers. In Ghana, women were recruited from 4 health facilities operating along a busy commercial corridor running through the Yilo Krobo and Lower Manya Krobo districts in the Eastern Region of the country. The livelihoods of the population in the catchment area were largely supported through petty trade, operating shops and kiosks, and providing skills and services. The RDNS study population was drawn from 11 rural unions of the Badarganj and Chirirbandar subdistricts of the northwest region of Bangladesh where the population was primarily engaged in farming, petty trade, transportation, and construction. The HFIAS survey instrument, which was developed in generic form by the Food and Nutrition Technical Assistance Project (6), was adapted to each local setting. After enrollment, the adapted questionnaire was administered 2 times to each household in the DOSE trial, 4 times during DYAD-M, 3 times during the DYAD-G trial, and 5 times during the RDNS trial. These self-reported food-insecurity data were collected at multiple time points to document the food-insecurity context for each household at each stage during the intervention, which spanned multiple seasons and, in some cases, multiple years. The adapted survey questions for each trial are available in Supplemental Methods 1. In the DOSE trial, the survey respondent was, in almost all cases, the mother or primary caregiver of the child enrolled in the trial. In the DYAD-M, DYAD-G, and RDNS trials, the survey respondent was the mother enrolled in the trial in almost all cases. For each of 9 food-insecurity access conditions, the survey respondent was asked whether anyone in her household had experienced the condition in the previous 4 wk. If yes, the respondent then indicated how frequently the condition occurred, where “rarely” was 1–2 times in the past 4 wk, “sometimes” was 3–10 times in the past 4 wk, and “often” was >10 times in the past 4 wk. The HFIAS score, a measure of the degree of food insecurity ranging from 0 to 27, was then calculated as the simple sum of the frequency-of-occurrence responses, where “never” was 0 points, “rarely” was 1 point, “sometimes” was 2 points, and “often” was 3 points. After the HFIAS questions were administered, respondents were then asked about strategies used to cope with food insecurity. The specific coping strategies were developed by using a subset of the generic strategies (23) and locally adapted through focus group discussions conducted at each site. The full text of the coping strategy questions, which were administered at each round of food-security data collection for the DOSE, DYAD-M, and DYAD-G trials and at 2 rounds of food-security data collection for RDNS, are available in Supplemental Methods 2. For the DOSE, DYAD-M, and DYAD-G trials (and to a much lesser extent for the RDNS trial), at each round of food-security data collection, there was substantial variation in the actual timing of data collection visits relative to when the visits were scheduled to occur. To compare food-security observations across households with a similar duration of exposure to the intervention in our analyses, instead of grouping food-security observation by round of data collection, observations were grouped by period, where each period represented a block of time relative to the age of the child enrolled in the trial (Table 2). Food-security data collection periods and sample sizes1 Women and children were randomly allocated to intervention groups across seasons during the rolling enrollment periods of each trial, but to account for possible imbalances across seasons in subsequent periods of food-security data collection, a seasonally adjusted HFIAS score was constructed. Seasons were identified using cropping calendars and personal communication with local contacts at each site, and seasons were defined as season by year (e.g., the lean season in 1 y was coded separately from the lean season in the following year) to allow for annual variation in seasonal food insecurity. With periods defined as in Table 2 corresponding to the child’s age, the seasonally adjusted HFIAS score for household i in season s and period p, was then defined in Equation 1 as: where was the average HFIAS score within the control group (IFA group in the case of the DYAD trials) in season s, and was the average HFIAS score within the control group (IFA group for DYAD trials) in period p. To preserve the integer nature of the score, seasonally adjusted HFIAS scores were rounded to the nearest integer, and negative scores were rounded to zero. The analyses were conducted by intent-to-treat and were performed separately for each trial. RDNS data from periods 1 and 2 were also analyzed separately from periods 3–5 because the combined intervention groups (described in Table 1) differed between the 2 sets of periods. Households with missed food-security visits were included in the analysis for all time points where data were available. In cases in which a food-security visit occurred far off schedule, resulting in 2 observations for the same household in one period, the visit closest to the scheduled date during that period was retained, and the other observation was dropped from the analysis. Analyses were conducted by using Stata 14 (StataCorp). The seasonally -adjusted HFIAS scores are essentially count data, and for all trials the distribution of scores was positively skewed. The effects of the DOSE, DYAD-M, and DYAD-G interventions on household food insecurity were therefore estimated by using negative binomial models with a household-level robust variance estimation to account for repeated measures. The RDNS models were estimated by using mixed-effect negative binomial models with random effects at 3 levels to account for the cluster design and the repeated measures: households nested within community health worker work areas, work areas nested within regional unions, and unions. All models included fixed effects for the period of food-security data collection. For the DOSE, DYAD-M, and DYAD-G trials, the scheduled baseline round of food-security data collection was done after random assignment for many, but not all, households, so the baseline round was omitted from those analyses. The baseline round was collected before random assignment for all RDNS households and was therefore included in all of the RDNS analyses as a covariate control. For all analyses, when the null hypothesis of no difference between intervention groups was rejected (P < 0.05), pairwise comparison incidence rate ratios, which are ratios of predicted seasonally adjusted HFIAS scores in the LNS intervention groups to the predicted scores in the non-LNS groups, were estimated and are referred to as predicted score ratios (PSRs). For analyses with >2 intervention groups, P values for post hoc pairwise comparisons of PSRs were adjusted for multiple comparisons by using Sidak’s method (24). Interaction terms between intervention group and period of data collection were used to assess differences in the effect of intervention group by period and were further examined by estimating the group marginal means for each period. Models that included an additional set of prespecified baseline covariates were estimated to determine whether adjusting for the additional covariates improved the precision of the estimated effects (25). Baseline covariates were included in the fully adjusted models if they were associated with the seasonally adjusted HFIAS score at the 10% level of significance in bivariate analyses. For all trials, season of food-security data collection, maternal age, and level of education and household electrification were included in fully adjusted models. For DOSE, maternal marital status and household distance to a main market were also included in fully adjusted models. DYAD-M and DYAD-G adjustment covariates also included maternal parity and household distance to a main market, and DYAD-M additionally included maternal marital status. In addition to the baseline seasonally adjusted HFIAS score, which was included in all RDNS models, RDNS fully adjusted models also included maternal parity and maternal BMI (in kg/m2). Effect modification by each baseline covariate was assessed by including a group-by-covariate interaction term. When seasonally adjusted HFIAS scores were significantly different between intervention groups, secondary analyses of individual HFIAS questions and food-insecurity coping strategies were performed to understand the drivers of the effect. Responses to individual HFIAS questions as well as responses to questions about specific food-insecurity coping strategies were coded as dichotomous variables and analyzed by using logistic models with household-level robust variance or mixed-effect logistic models with random effects of union, cluster, and household.