Drivers of food consumption among overweight mother-child dyads in Malawi

To address the increase in overweight and obesity among mothers and children in sub- Saharan Africa, an understanding of the factors that drive their food consumption is needed. We hypothesized food consumption in Malawi is driven by a combination of factors, including season, food accessibility (area of residence, convenience of purchasing food, female autonomy), food affordability (household resources, food expenditures, household food insecurity), food desirability (taste preferences, body size preferences), demographics, and morbidity. Participants in Lilongwe and Kasungu Districts were enrolled across three types of mother-child dyads: either the mother (n = 120), child (n = 80), or both (n = 74) were overweight. Seven-day dietary intake was assessed using a quantitative food frequency questionnaire during the dry and rainy seasons. Drivers associated with intake of calories, macronutrients, and 11 food groups at p<0.1 in univariate models were entered into separate multivariate linear regression models for each dietary intake outcome. Mother-child dyads with an overweight child had a higher percent of calories from carbohydrates and lower percent of calories from fat compared to dyads with a normal weight child (both p<0.01). These mothers also had the highest intake of grains (p<0.01) and their children had the lowest intake of oil/fat (p = 0.01). Household food insecurity, maternal taste preferences, and maternal body size preferences were the most consistent predictors of food group consumption. Household food insecurity was associated with lower intake of grains, fruits, meat and eggs, oil/fat, and snacks. Maternal taste preferences predicted increased consumption of grains, legumes/nuts, vegetables, fish, and oil/fat. Maternal body size preferences for herself and her child were associated with consumption of grains, legumes/nuts, dairy, and sweets. Predictors of food consumption varied by season, across food groups, and for mothers and children. In conclusion, indicators of food affordability and desirability were the most common predictors of food consumption among overweight mother-child dyads in Malawi. Copyright: The study was conducted in Lilongwe and Kasungu Districts in the Central Region of Malawi, which had the highest prevalence of overweight among women and children in the most recent Demographic and Health Survey [2]. The data collection took place in both the dry season (May-October 2017 and 2018) and rainy season (January-April 2018 and 2019), on average, 6.6 ±1.0 months (range: 4.1 to 10.7 months) later. In each district, data were collected in two urban neighborhoods and two rural villages. Local leaders in the study sites invited all women with children less than 5 years for anthropometric screening to determine study eligibility. Standing height of mothers and children ≥2 years was measured to the nearest 0.1 cm using a portable stadiometer (Seca 213). Recumbent length of children <2 years was measured to the nearest 0.1 cm using an infant measuring mat (Seca 210). Weight of mothers and children was measured to the nearest 0.1 kg using a digital scale (Seca 803 for mothers and children ≥2 years; Seca 354 for children <2 years). Three types of mother-child dyads were purposefully enrolled based on their nutritional status as determined during the screening: (1) overweight mothers (body mass index [BMI] ≥25 kg/m2) with an overweight child (weight-for-height z-score [WHZ] >+2 SD), (2) overweight mothers with a normal weight child (WHZ> -2 SD and WHZ ≤+2 SD), and (3) normal weight mothers (BMI ≥18 kg/m2 and BMI <25 kg/m2) with an overweight child. The sample sizes for the mother-child dyads were as follows: overweight mother, overweight child (n = 74); overweight mother, normal weight child (n = 120); and normal weight mother, overweight child (n = 80). Additional eligibility criteria included: mothers aged ≥18 years, children aged 6–59 months, and mothers were the biological parent of the enrolled child. Data on pregnancy status were not collected. The study was approved by the College of Medicine Research Ethics Committee at the University of Malawi and by the institutional review boards at RTI International and the Harvard T.H. Chan School of Public Health. Participants provided signed informed consent. They received an incentive equivalent to $4 USD. All data were collected by trained research assistants in the participants’ local language (Chichewa). Quantitative food frequency questionnaires (QFFQs) adapted from a previous study in South Africa [13–15] were used to assess habitual dietary intake of mothers and children. The QFFQ included food flash cards (photos) for all foods based on lists from existing Malawi food composition tables [10, 16] and scans of food items available at local stores and markets. Data were collected on diet recalled for the previous 7 days. The mother was asked to create one pile of food cards showing items she rarely (once every few months) or never ate or drank and then to divide the remaining cards into items she ate or drank occasionally (once or twice a month) and those she ate regularly (every day or every week) [17]. She was then asked for information on the frequency and amounts of the regular food items, repeating the same process for the child. Intake of cooking oil or butter/margarine of 50 grams per day or more was divided by household size as research assistants indicated that these values were reported at the level of the household rather than the individual. Eleven food groups were derived from the data to evaluate as outcomes, separately for mothers and children: grains, roots/tubers, vegetables, fruits, meat/eggs, fish, dairy, legumes and nuts, oil/fat, snacks, and sweets (S1 Table). Total energy (kcal/day) and percent of energy from carbohydrates, fat, and protein were determined using a Malawi food composition table [18]. Nutrient composition for 37 missing food items was obtained from the Tanzania food composition table [19] or the USDA food composition database [20]. Drivers of food consumption specified a priori and their link to the food environment conceptual framework are summarized in Table 1 [12]. We also included several other individual and household level factors that have been identified in previous studies of drivers of dietary behaviors in SSA [21]. Access to food relates to physical location, distance to food retailers, mode of transport, and who has the responsibility to purchase food for the household. Specific variables considered in this category were residence in Lilongwe (referent) versus Kasungu, urban (referent) versus rural residence, how the woman gets to the nearest market/shop to purchase food (walk [referent] or other), and how long it takes to get to nearest market/shop to purchase food (minutes). We also included who purchases most food for the household (mother [referent], husband/partner, both, or other family member) and female autonomy score in the food access category. A validated multidimensional construct was used to measure female autonomy. Mothers were asked eight questions on freedom from violence, participation in non-economic family decisions, community involvement, and participation in household economic decisions, with a score of one being recorded for each question where decisions were made herself or jointly with a husband or partner while a score of zero was assigned in all other cases [22]. For the final question, which asked if respondents thought a husband is justified in hitting or beating his wife in each of five situations, 0.2 points were given for each time that the mother reported that the husband was not justified. Points given for all questions were then summed to create the female autonomy score, with a higher score indicating higher autonomy. A questionnaire adapted from a Feed the Future evaluation in Malawi was used to determine household assets and expenditures on food [23]. Food expenditure was recalled over the past 7 days and the total amount spent on food for the household in a typical week was calculated in local currency and converted to USD using the World Bank official exchange rate for 2018 of 732.33 (accessed October 1, 2019). This was also how total amount spent on special foods for children <5 years in the household was calculated (USD converted using World Bank official exchange rate for 2018 of 732.33 [accessed October 1, 2019]). Proxies for purchasing power included main source of drinking water (piped [referent], well, or borehole), toilet facility used by the household (flush [referent], ventilated improved latrine, pit latrine with roof, or traditional pit latrine/no facility), total number of household assets, and household food insecurity. A household asset score was calculated as the total number of the following assets owned by the household (maximum score of 12): electricity, koloboyi (home-made kerosene lamp), paraffin lamp, radio, television, mattress, sofa set, table and chair(s), refrigerator, watch, bicycle, and mobile telephone [2]. The Household Food Insecurity Access Scale (HFIAS) questionnaire was used to assess household food insecurity. Participants were asked to recall how frequently over the past four weeks each of nine conditions was experienced, covering the following three domains: (i) anxiety and uncertainty, (ii) insufficient quality, and (iii) insufficient food intake and its physical consequences [24]. Responses were used to calculate an HFIAS score with a minimum of 0, if the household responded “no” to the occurrence of all conditions, and a maximum of 27, if the household responded “yes” to the occurrence of all conditions with a frequency of “often” for all conditions. Thus, higher HFIAS scores corresponded to greater food insecurity [24]. This category included variables related to preferences, tastes, desires, attitudes, and culture. Specific variables included maternal taste preference for the corresponding food group (e.g., preference for grains as a predictor of grain intake), mother’s body size preference for herself and her child (underweight [referent], normal weight, overweight, or obese), mother’s perception of a healthy body size for herself and her child (underweight [referent], normal weight, overweight, or obese), and purchasing special foods for children <5 years in household (yes or no [referent]). To assess maternal taste preferences, we used a taste preference checklist matching items in the QFFQ and representing all major foods consumed in this population [9, 10, 16]. Participants indicated how much they liked or disliked each item using a 5-point hedonic preference scale, ranging from “1 = extremely dislike” to “5 = extremely like” with “3 = neither like nor dislike” [25, 26]. Preferences for the 11 food groups were calculated as the arithmetic average preference for all foods in that group. To assess body size preferences, a set of seven adult female and seven child body silhouette drawings were used. A local artist adapted mothers’ body silhouettes from versions previously used in Malawi [27–29] and adapted child body silhouettes from Hager et al. [29]. Each body silhouette drawing was printed separately on cardstock and laminated. The interviewer mixed the body silhouettes and laid them out in a random order before asking the mother to make selections for herself and then again for her child [30]. Demographic characteristics included the age of the mother (years), the age of the child (months), child sex male (referent) versus female, household size, number of children <5 years in the household, and maternal educational attainment less than secondary school (referent) versus any secondary school or higher. We also considered maternal and child morbidity over the past two weeks (yes or no [referent] to diarrhea, fever, and cough) as a driver. Morbidity questions for children were from the Malawi Demographic and Health Survey and the same questions were used for mothers [31]. All drivers were surveyed in both seasons except for mother’s body size perceptions and preferences for herself and the child, which were measured only in the dry season. We did not repeat anthropometric measurements in the rainy season because the analysis was based on the body size of the mother-child dyads at enrollment. Descriptive statistics were used to summarize the hypothesized drivers and dietary intake. Differences across the three types of mother-child dyads were determined using chi-square tests and Kruskal-Wallis H tests (a non-parametric alternative to one-way analysis of variance). Differences between seasons were tested using exact tests of symmetry (which reduce to McNemar’s tests in the case of binary variables) [32] and Wilcoxon signed-rank tests (a non-parametric alternative to a paired t-test). The relationship between each driver (independent variables) and each of the four nutrient variables (total calories and percent of calories from carbohydrates, fat, and protein) and 11 food groups (dependent variables) was evaluated using linear regression. Drivers associated with the outcome at p<0.1 in univariate linear regression models (e.g. models with one independent variable) were entered into a multivariate linear regression model (e.g. models with more than one independent variable; 15 separate models, one for each of the dietary intake outcomes) and relationships significant at p<0.05 were presented in the results. Continuous drivers were rescaled to have a mean of zero and a standard deviation of one to ease interpretation and comparisons across variables. All analyses were conducted using Stata v. 14.2 (StataCorp, College Station, Texas, US).