Standard Minimum Dietary Diversity Indicators for Women or Infants and Young Children Are Good Predictors of Adequate Micronutrient Intakes in 24-59-Month-Old Children and Their Nonpregnant Nonbreastfeeding Mothers in Rural Burkina Faso

Background: Simple proxy indicators are needed to assess and monitor micronutrient intake adequacy of vulnerable populations. Standard dichotomous indicators exist for nonpregnant women of reproductive age and 6-23-mo-old children in low-income countries, but not for 24-59-mo-old children or pregnant or breastfeeding women. Objectives: This study aimed to evaluate the performance of 2 standard food group scores (FGSs) and related dichotomous indicators to predict micronutrient adequacy of the diet of rural Burkinabe 24-59-mo-old children and women of reproductive age by physiological status. Methods: A 24-h recall survey was conducted at dry season among 1066 pairs of children and caregivers. Micronutrient adequacy was evaluated by the mean probability of adequacy (MPA) of intake over 11 micronutrients. Proxy indicators were FGS-10 [10 food groups based on the FAO/FHI360 minimum dietary diversity for women (MDD-W) guidelines] and related MDD-W (FGS-10 ≥5); and FGS-7 [7 groups based on the WHO infant and young child (IYC) feeding MDD guidelines] and related MDD-IYC (FGS-7 ≥4). Results: FGS-10 and FGS-7 were similar across children and women (∼3 groups). FGS-10 performed better than FGS-7 to predict MPA in children (Spearman rank correlation = 0.59 compared with 0.50) and women of all 3 physiological statuses (Spearman rank correlation = 0.53-0.55 compared with 0.42-0.52). MDD-W and MDD-IYC performed well in predicting MPA >0.75 in children and MPA >0.6 in nonpregnant nonbreastfeeding (NPNB) women, but a 4-group cutoff for FGS-10 allowed a better balance between sensitivity, specificity, and proportion of correct classification. MPA levels for pregnant and breastfeeding women were too low to assess best cutoff points. Conclusions: MDD-IYC or an adapted MDD-W (FGS-10 ≥4 instead of FGS-10 ≥5) can be extended to 24-59-mo-old children and NPNB women in similar-diet settings. The inadequacy of micronutrient intakes in pregnant and breastfeeding women warrants urgent action. Micronutrient adequacy predictors should be validated in populations where a higher proportion of these women do meet dietary requirements. Burkina Faso is a low-income food-deficit country located in West Africa (14). The 2017 national nutritional surveillance survey found that 21% of children aged <5 y were stunted (height-for-age z-score < −2), and 9% were wasted (weight-for-height z-score < −2), and highlighted particularly poor infant and young child feeding (IYCF) practices (15). This survey also showed that, in 2017, only 18% of the Burkinabe children and 20% of women of reproductive age met the minimum dietary diversity (defined as having consumed 4 of 7 food groups for children and having consumed 5 of 10 food groups for women of reproductive age). This study draws on data collected as part of the SELEVER (Soutenir l'Exploitation familiale pour Lancer l’Élevage des Volailles et Valoriser l’Économie Rurale) impact evaluation, including 2 rounds of surveys (baseline and 6-mo follow-up) undertaken prior to the intervention roll-out. The detailed SELEVER trial protocol has been published elsewhere (16). Briefly, the SELEVER cluster randomized control trial is an ongoing 5-y study evaluating the impact of an integrated agriculture-nutrition package of interventions on the diets, health, and nutritional status of women and children in Burkina Faso. The SELEVER study population includes 120 rural communities across 3 regions of Burkina Faso: Boucle de Mouhoun, Centre-Ouest, and Haut-Bassins. These regions were prioritized for the SELEVER roll-out based on the potential for the poultry production sector to meet the demand from urban markets (17). During the preparation stages of the SELEVER trial, 60 communes were selected from a pool of 79 communes available for the scale-up of the intervention (Supplemental Figure 1). The primary outcomes of the SELEVER trial are: 1) MPA of diets for women and children (2–4 y at baseline) measured through dietary assessment using the interactive 24-h recall multiple-pass method at baseline, and after 6, 30, 36, and 48 mo; 2) IYCF practices for children aged 0–2 y measured using interviews with caregiver at baseline, and after 6, 30, 36, and 48 mo; and 3) household poultry production (output and sales) measured using interviews with caregiver at baseline, and after 6, 30, 36, and 48 mo. The 2 rounds of data were collected to allow for measurement of preintervention seasonal variations across the primary outcomes of the trial, with surveys timed during the dry and rainy seasons. The first survey was conducted from March to June 2017 in the dry season just before the first rains and prior to the main planting period. The follow-up survey was undertaken from September to October 2017, during the preharvest rainy season period. Both survey rounds included data collection at child, caregiver, household, and village level. The main analyses in this article use data from the dry season; data from the rainy season were used to assess the robustness of results and are presented as supplemental tables. This study analyzes dietary intake data of children aged 24–59 mo at the time of the baseline survey and of their primary female caregiver. The sampling and household selection protocol has been described in detail elsewhere (16). It involved a 2-stage sampling to accommodate budget constraints and different sample size calculations depending on the outcome of interest. The sample for dietary outcomes was part of the second stage of randomization (Supplemental Figure 1). Briefly, prior to the first baseline survey, a household census was conducted to identify all households with children in the 24–59-mo age group, and all the large poultry-producing households (defined as owning a poultry flock of >20 chickens/fowls) in the community. Twelve households were then randomly selected in 90 of the 120 rural communities for the dietary survey interviews, stratifying by poultry flock size status. Within each sampled household, an index child was randomly selected among all children aged between 24 and 59 mo. The caregiver could be the biological mother of the index child or another woman taking care of the child in the household. Data were collected from a sample of 1066 pairs of caregivers and children (PCCs) enrolled at the baseline survey (Supplemental Figure 2). Collected data included dwelling features, sociodemographic characteristics of household members, food security information using the household food insecurity access scale classification (18), and dietary intakes of the PCC. The study received ethical approval from the Comité d’Ethique pour la Recherche en Santé (national ethic comitee, Ministry of health/ Ministry of research) in Burkina Faso and the International Food Policy Research Institute IRB (Institutional Review Board) in Washington, DC. All caregivers provided written informed consent for themselves and their child as well. Dietary intake data were collected using an interactive 24-h recall method (19). Caregivers were asked to report everything they and their child had eaten the previous day. After compiling a list of all foods that were eaten, a questionnaire was filled out using computer-assisted personal interviewing with information for each individual food item, including quantities consumed, quantities of leftovers, and detailed recipes for composite dishes. To facilitate the quantification of portion sizes and to avoid sharing food from a common pot, enumerators distributed 2 bowls and 2 plates to the caregiver 2 d before the interview and advised them to serve themselves and their index child all meals using these bowls or plates until the 24-h recall would take place. During the recall, portion sizes were estimated using the distributed plates and bowls and other common household measures, water volume, images, or clay or wooden models, if it was not possible to directly weigh a food replicate. During the follow-up survey, data were collected from the same PCC using the same 24-h recall method. To capture the intraindividual variation of intakes necessary to estimate usual intakes, a second 24-h recall was undertaken on a nonconsecutive day in ∼16% of the sample during both survey rounds. Food quantities were collected in grams, volume, prices paid for purchase, or in terms of household measures units and later converted to grams using context-specific conversion lists. Several of the context-specific conversion factors and all price-related conversion factors were collected during market surveys conducted concurrently with the household surveys. Food quantity values obtained were converted into nutrients, using either the FAO West African food composition table (FCT) (20) or another FCT adapted to Burkinabe foods (21), after adjustment for micronutrient retention factors for cooked foods (22) and edible portions. All individuals were included in the analysis except for 44 women (4% of the sample) who reported fasting for Lent (which was not common in our population), and 1 woman who fasted for other reasons and consumed only 1 cup of coffee during the previous day (Supplemental Figure 2). As recommended (23), to avoid introducing an unknown bias, we included in the analysis the overreporters (16% in women, 3% in children) and underreporters (11% in women, 8% in children) according to the Goldberg cutoff method (24). The micronutrient adequacy of individuals was assessed for 11 micronutrients: vitamin A, vitamin C, thiamin, riboflavin, niacin, vitamin B-6, vitamin B-12, folate, calcium, zinc, and iron. Vitamin A was expressed in micrograms of retinol activity equivalents and other micronutrients in micrograms or milligrams. The probability of adequacy (PA) for each micronutrient of interest was assessed through the probability approach (25) using the estimated average requirements (EARs) and SDs for women aged <18 y, women aged 19–49 y (pregnant, breastfeeding, or NPNB), and children aged 24–59 mo, based on WHO/FAO recommendations (26). For iron, whose distribution is generally skewed, we used the Institute of Medicine approach to calculate the PAs (27), accounting for the low bioavailability of iron (assumed to be 5%) in our context. We used the recommendations issued by the International Zinc Nutrition Consultative Group to calculate the PA for zinc, considering the lowest level of bioavailability (28). We used Institute of Medicine recommendations to calculate the PA for calcium (29). For the remaining nutrients (niacin, thiamin, riboflavin, folate, and vitamins A, C, B-6, and B-12), a Box–Cox transformation was used to obtain symmetrical distributions and, then, using the intraindividual and interindividual variances obtained from the second recall on a nonconsecutive day conducted in 16% of the households, the best linear unbiased predictor of an individual's usual intake was calculated (30). Depending on the relative size of the intraindividual to the interindividual variability in intake for each nutrient, the best linear unbiased predictor shrinks the person-level mean intake toward the overall group mean. The PA was then estimated for each micronutrient as the probability that an individual's usual intake is above the actual requirement for that micronutrient, the latter being along the normal distribution of requirements, with known EAR and SDs (25). However, for vitamin B-12, because the majority of zero values in its distribution did not allow the calculation of usual intakes, the actual intakes were used to estimate the PA. For each individual micronutrient, the mean PA at population level corresponds to the prevalence of individuals in that population covering their dietary requirements for the micronutrient. An overall MPA was then calculated for each individual by averaging the PA values across the 11 micronutrients considered. The 24-h dietary recall data were used to derive FGSs. Each individual food was classified into a food group. Two validated classifications commonly used in the literature were used to build 2 FGSs. The first FGS (named FGS-10) was based on the minimum dietary diversity for women (MDD-W) guidelines whose food groups are the following: 1) grains, white roots and tubers, and plantains; 2) pulses; 3) nuts and seeds; 4) dairy; 5) flesh foods; 6) eggs; 7) dark-green leafy vegetables; 8) vitamin A–rich fruits and vegetables; 9) other vegetables; and 10) other fruits (8). This classification was shown to provide an FGS that is an adequate proxy of the MPA of nonpregnant women (9). The second FGS (FGS-7) was based on the IYCF minimum dietary diversity (MDD-IYC) guidelines and had the following groups: 1) grains, roots, and tubers; 2) legumes and nuts; 3) dairy products; 4) flesh foods; 5) eggs; 6) vitamin A–rich fruits and vegetables; and 7) other fruits and vegetables (7). Oils (except vitamin A–enriched oils and red palm oil, which were classified among vitamin A–rich fruits and vegetables), drinks, and condiments were not classified into food groups. Foods whose total consumption during the day was <10 g were considered condiments and not included in the food groupings (31). We applied an identical analytical strategy to the sample of children and the 3 subsamples of caregivers: 1) pregnant women; 2) breastfeeding women; and 3) NPNB women. First, we assessed the association between the discrete FGS-10 or FGS-7 and the MPA. Given the skewed distribution of the MPA distribution for both women and children, associations between the MPA and the 2 FGSs were assessed using Spearman rank correlation coefficients, both with and without adjustment for energy intake (32). Then, the second analysis aimed to assess the performance of the FGS indicators in identifying individuals with low or higher MPA. For this purpose, the area under the receiver operating characteristic curve (AUC) values was calculated for the 2 highest possible cutoffs for MPA (including >0.6;  >0.7; >0.75;  >0.8; or >0.9) with ≥50 individuals above the cutoff points. Finally, sensitivity, specificity, and percentage of correct classification were calculated to assess the performance of various FGS cutoffs to screen for “higher” MPA, defined as >0.75 for children and >0.6 for NPNB women. This definition of a “higher” MPA is based on the distribution of MPA in the population and the cutoffs selected in previous studies for young children and NPNB women (4, 9, 10, 13). Sensitivity was defined as the percentage of children (or women) who had an FGS greater than a prespecified cutoff (e.g., 4 groups) among children (or women) with an MPA greater than the prespecified desired level (e.g., 0.6). Specificity was defined as the percentage of children (or women) whose FGS was lower than the cutoff considered among children (or women) who did not reach the required MPA threshold. These analyses were not carried out for pregnant or breastfeeding women because there were not enough women reaching a MPA >0.6. Robustness analyses were also conducted using various levels to define adequate MPA. Because of the wide range of PA contributing to the MPA, we assessed the association between FGS and the PA for each micronutrient. We also assessed the performance (sensitivity, specificity, and percentage of correct classification) of the optimal FGS cutoff, as determined by the MPA analysis, to predict adequate micronutrient intake, defined as individual micronutrient PA >0.8. We also repeated the whole analysis using the lean season data from the same PCC to assess the robustness of the findings. In addition, using both rounds, we used linear mixed-effects regression models to confirm the overall performance of the 2 FGSs to predict MPA. The models included random effects at the village and individual (child or woman) level, and fixed effects for the season and for child sex and age, maternal breastfeeding status, child or woman energy intake, and woman’s age. The interaction of FGSs with the season was also tested. The assumption of normality of residuals from those regressions was assessed using the Shapiro–Wilk test (33). Finally, other robustness analyses were conducted without considering a minimum threshold to count a group.