Nutrition-specific and nutrition-sensitive factors associated with mid-upper arm circumference as a measure of nutritional status in pregnant Ethiopian women: Implications for programming in the first 1000 days

Poor nutritional status in pregnancy expressed as low mid-upper arm circumference (MUAC) is associated with low birth weight. The study aims were to assess the nutritional status of pregnant Ethiopian women using MUAC and examine association with nutrition-specific and nutrition-sensitive factors, using baseline data of a prospective longitudinal observational birth cohort study conducted in three rural districts in the Oromia region of Ethiopia. Recruitment into the cohort was rolling over a period of nine months, and the data used for this analysis were collected while the women were between 12–32 weeks of gestation. Detailed household socio-demographics, agricultural production, women’s health, morbidity and diets, with weights, heights and MUAC, and anemia prevalence (HemoCue) were collected. The prevalence of low MUAC (< 23 cm) was 41%. Controlling for location and clustering, wealth quintile (OR = 0.88, CI = 0.82 to 0.96, p<0.01) was associated with decreased risk of low MUAC, while trimester (OR = 1.31, CI = 1.16 to 1.48, p<0.001) was associated with increased risk of low MUAC. The only significant factor amenable to nutrition-specific interventions was altitude-adjusted anemia, which was associated with increased risk of low MUAC (OR = 1.28, CI = 1.09 to 1.49, p<0.01). Significant factors amenable to nutrition-sensitive factors and associated with higher odds of low MUAC were household food insecurity (OR = 1.04, CI = 1.02 to 1.06, p<0.001), distance to the clinic in minutes (OR = 1.01, CI = 1.0 to 1.01, p<0.0001) and season of recruitment (lean versus non lean) (OR = 1.30, CI = 1.10 to 1.54, p<0.01). Literacy (OR = 0.85, CI = 0.74 to 0.98, p<0.05) and numeracy (OR = 0.75, CI = 0.62 to 0.91, p<0.01) were also significantly associated with lower odds of low MUAC. Poor nutritional status in pregnancy expressed as percent with low MUAC was high in Ethiopian women. It was associated with several nutrition-specific and -sensitive factors indicating the importance of multisectoral actions in improving outcomes within the first 1000 days. The study population consists of pregnant women recruited as part of a longitudinal birth cohort study (implemented from 2014 to 2016) conducted in the three districts (called woredas) of Woliso, Goma and Tiro Afeta in the Oromia region of Ethiopia. The overall aim of the longitudinal study was to assess the effectiveness on maternal and child nutrition outcomes of an integrated nutrition/agriculture program (Empowering New Generations in Nutrition and Education—ENGINE) supported by the United States International Agency for Development (USAID). The study had an open cohort design and allowed rolling enrollment of pregnant women identified by health workers at the community level and at health posts from March 2014 to January 2015. The sample size was estimated to detect a 0.14 (1 standard-deviation) change in length-for-age z-score from birth to 12 months of age (in infants born to the recruited women), at 80% power at the 0.05 level of significance, allowing for 30% attrition. Length-for-age z-score was used as the measure for sample size computations as the programmatic intervention aimed to reduce the prevalence of stunting. Recruitment was conducted at the lowest administrative cluster level within the district (called a kebele). Using an intra-cluster correlation of 0.03, a total of 117 clusters (kebeles) with a total sample size of 4680 (n = 40 per cluster) was estimated. All kebeles within a district were sampled–with the exception of a few excluded due to inaccessibility–for a total of 1,560 pregnant women recruited in each of the three districts. The study received ethical approval from Jimma University in Ethiopia (approval reference number RPGC/264/2013) and from Tufts University in the United States (IRB reference number 11088). Informed consent was obtained for every woman recruited in this study for themselves and their unborn infant. All data were collected in assessments administered at the household. Following receipt of informed consent, women were administered a urine test to confirm pregnancy, were tested for malaria using a Rapid Diagnostic tests (RDT) (Malascan) and also tested for anemia using HemoCue to determine inclusion. Exclusions and referrals were made for pregnant women who were found to be severely malnourished (MUAC < 18.5), with severe anemia (< 7 g/dl hemoglobin) or malaria, or with pregnancy induced hypertension. Multiple births were excluded. The age range was 15 to 50 years old with gestation age between 12 to 32 weeks. A significant proportion of married women within these districts were below 18 years of age and it was considered necessary to include these women for routine follow ups. Both IRB committees provided approval for recruiting women under 18 years of age as, under Ethiopian law, married women (irrespective of age) are considered as emancipated. Data collection started in pregnancy and ended at 12 months postpartum. Data were collected in pregnancy, at the infant’s birth, and every 3 months until 12 months postpartum. Data on maternal and household characteristics were collected at every time point whereas infant data were collected at birth and across four time points until the end of the study. The data used in this analysis are from the pregnancy timepoint of this longitudinal study. These include pregnancy data, household characteristics, access to services, diet and food security. The methods for the overall study are further described in another publication by the same research group [22]. The data for this paper are limited to a cross-sectional round of the longitudinal study, i.e., the baseline/enrollment timepoint during pregnancy. A total of 4,560 women had complete data that are utilized in the analysis included in this paper. Explanatory variables identified through literature are described in Table 1 and include individual, household, and district-level agricultural and service related variables. The dependent variable is prevalence of low MUAC which is defined as the percentage of women with MUAC as less than 23 cm at the time of study enrollment [23]. Ɨ denotes median and IQR While other factors within the domains of access and use of services and agricultural characteristics improved the model, with the exception of health clinic distance, none of the other variables increased nor decreased significantly the odds of a low MUAC. Explanatory variables that were computed and tested against the dependent variable were classified into the domains of maternal characteristics (trimester, time of recruitment, age, literacy, numeracy, anemia, trimester, months since last pregnancy, number of prior pregnancies, minimum dietary diversity score, hand washing score), household characteristics (household food insecurity access score, household size, household religion, type of household- female headed, wealth quintile, access to protected water source), agricultural characteristics (crop production diversity score, livestock ownership diversity score,) and access and use of services (market distance, health clinic distance, health worker home visits, and number of visits to a health facility in the past year). These variables are amenable to both nutrition-specific and nutrition-sensitive interventions. Control variables included location and clustering at the kebele level. Women are more likely to have poor nutritional status (as reflected by the low MUAC) in the lean season [24]. Women were asked the number of months that their households were unable to meet food needs according to the questions for the Months of Adequate Food Provisioning (MAHFP) [25]. We found that across the period of 12 months, 40–80% of households were unable to meet food needs from June through September. The rest of the year the percentage of households not meeting food need ranged from 1–18%. Assuming these as the most likely months as lean for the households and subsequently to the pregnant woman, we created a binary variable to indicate recruitment/data collection in the lean months (June-September) compared to recruitment in the non-lean season (October-May). Either age or date of birth was reported by the woman at recruitment, and age was calculated (if necessary). Weight was estimated as an average of three measurements using a SECA measuring scale with a precision to the nearest 100g. Height was an average of three measurements using a Standiometer wooden measuring board, with a precision of 0.1 cm. MUAC was derived as an average of three measurements to the nearest millimeter, using a flexible non-elastic tape, midway between the tip of the shoulder and the tip of the elbow of the left arm with the arm hanging freely by the woman’s side. Literacy and numeracy variables were computed for each participant. Women were classified as literate if they were able to read specific sentences in Amharic and numerate if they correctly answered a simple math problem [26]. Prevalence of anemia (altitude-adjusted) and the mean hemoglobin levels were estimated. Hemoglobin levels were measured with the HemoCue system for mobile screening. Hemoglobin cutoff values were adjusted for altitude (GPS) and trimester according to method described by Cohen and Hass [27]. Adjusted hemoglobin cutoffs produced using this method range from 104.8 to 130.5 g/L. The average kebele altitude was imputed for missing altitude values. Women with hemoglobin levels below the adjusted cutoff point were classified as anemic. The LMP (last menstrual period) method was utilized to ascertain the gestational age/trimester of recruitment. Enumerators asked the women for the date of the last menstrual period, which was used to estimate the number of weeks pregnant. Trimester cutoffs were coded as 12 weeks and below for first, 13–27 weeks for second, and 28 and above for third trimester [28]. In order to capture birth spacing, women were asked how many months since their last pregnancy. Because recommended birth spacing is at least 24 months [29], we created a binary variable to capture women who gave birth within the previous 24 months. Women were asked the number of previous pregnancies, excluding the current pregnancy, they had and this number was used as a continuous measure in the model. Women’s dietary diversity was estimated using the food group categories outlined by FAO for computing the Minimum Dietary Diversity Score for Women (MDD-W) [30]. The MDD-W is a proxy indicator for nutrient adequacy of the diet and was derived from a qualitative 24-hour dietary recall. While we used the method to determine food groups and a count of food groups, we did not compute the MDD-W preferring to utilize the score as a continuous variable. The multi-pass method was used for the 24-hour recall: first, all foods and drinks consumed during a 24-hour period were listed; next, amount consumed using portion size estimation with models; thirdly, more details on the food (color, type, size, brand) were gathered; and finally, all information was reviewed and verified. Dietary data was coded in Excel and grouped into the following categories: 1) all starchy staple foods, 2) beans and peas, 3) nuts and seeds, 4) dairy, 5) flesh foods, 6) eggs, 7) vitamin A-rich dark green leafy vegetables, 8) other vitamin-A-rich vegetables and fruits, 9) other vegetables, 10) other fruits. Due to very low dietary diversity, the continuous score was used instead of the minimum cut point of 5 out of 10 food groups [30]. A hand-washing score was calculated based on a count of seven responses to questions about important times for hand-washing (when dirt is visible, prior to eating, before food preparation, before serving a meal, before feeding a child, after using the toilet, after cleaning a child following defecation). Household food insecurity access scale (HFIAS) score was constructed using the method described by Coates et al 2007 [31]. The woman was asked whether a series of nine conditions related to food security occurred in the previous four weeks (yes or no), and if yes, how frequently the condition occurred (rarely, sometimes, or often). In order to create the HFIAS continuous score, the frequency-of-occurrence responses were coded to 0 for each condition that did not occur, and frequency-of-occurrence responses were summed [31]. Household size was estimated by the number of members of all ages recorded in the household roster as reported by the index woman. A wealth index was created using polychoric principal component analysis. Variables included in the wealth index estimation were main type of walls, roof, floor, toilet, cooking fuel, source of energy, drinking water source, other household water source, and water treatment method [32]. The wealth index was then divided into quintiles. In order to capture water quality, households reporting their most common source of water to be piped, public tap, tube well or borehole, protected well or spring, or rainwater were coded as using a protected water source; households reporting other water sources such as an unprotected well or spring, river, or pond were coded as using an unprotected water source [33]. Crop production diversity was calculated as a simple count of the crop groups produced annually by the household [19]. Crops were grouped as cereals, roots and tubers, legumes, cash crops, vegetables, fruits, oil seeds, and spices for a maximum score of 8. A livestock ownership diversity score was generated as a count of the types of livestock that households reported currently owning within the following categories: 1) cattle, 2) camel, 3) sheep, 4) goat, 5) donkey, 6) horse, 7) mule, 8) chicken, 9) other poultry, 10) bee 11) other livestock. Time in minutes to the nearest major and local markets were reported by the household head. The market distance variable was constructed by time in minutes to the nearest of the two (major or local) for each household. Access to health clinic is measured by the time in minutes to nearest health clinic/post. Healthcare service usage was measured by asking women if they had visited a health facility and/or if they had received a home visit by a health worker in the past year. Two separate binary variables were created, with women reporting having visited a health facility in the past year, and women reporting receiving a home visit by a health worker in the past year. Descriptive statistics were generated that allowed for finalization of the variables to be included in the final model. These included generating means and standard deviations and prevalence estimates and frequencies. All analyses were conducted using StataSE 14 software. Tests of normality were conducted using the Shapiro-Wilk test. All explanatory variables were tested for associations with the dependent variable (low MUAC) and for associations with each other. For bivariate comparisons of means, we used Student’s t-test, and for comparison of categorical variables, chi-square tests were used. Explanatory variables that were associated with low MUAC at the bivariate level but were non-significant, duplicative or collinear with other explanatory variables were discarded. Multivariate logistic regression analyses included woreda fixed effects to control for characteristics common by region and standard errors were clustered at the kebele level using the vce command in STATA. Model fit tests were conducted.