Gender and household structure factors associated with maternal and child undernutrition in rural communities in Ethiopia

Addressing maternal and child undernutrition is a priority for the National Nutrition Program of Ethiopia. In a cross-sectional design, we selected mother-child pairs (n = 630) from Halaba, south Ethiopia (n = 413, two communities) and Zeway, Oromiya region (n = 217, one community). These communities were previously included in a project to improve agricultural practices. We aimed to estimate the level of maternal and child undernutrition in the two study sites and compare findings to regional/national reports. We also examined associations with gender, household-structure and nutrition/health related variables. Households were selected using simple random sampling based on list of households obtained from local health posts. Mothers were interviewed via questionnaire. Anthropometric measurements were taken from mothers-child pairs. Maternal undernutrition (% BMI<18.5) ranged from moderate (14% Zeway) to high (22% Halaba). In the children, stunting and underweight were very high (54% and 42% stunting, 36% and 21% underweight, in Halaba and Zeway, respectively). Up to 95% of Halaba and 85% of Zeway mothers reported 'same as usual' or 'less than usual' consumption patterns during their most recent pregnancy compared to periods of non-pregnancy. Mothers reported (61% in Halaba, 18% in Zeway) abstaining from consumption of certain nutritious foods for cultural reasons. Gender and socio-economic-demographic structure of the households, including imbalance of power, control of farm produce, physiological density, household size and dietary habits during pregnancy showed significant associations with maternal and child undernutrition (p<0.05). The levels of child and maternal undernutrition, particularly in children, were unexpected and of concern, given that a national nutrition program has been in place since 2008. The study provides insights for policy makers to improve women's education, reproductive health services for better family planning, and strengthen nutrition/health programs designed to target vulnerable segments of the population in these and other rural communities and districts with similar structure and demographics in Ethiopia. The baseline data reported here was collected as part of a community-based intervention study in three purposively selected rural Kebeles (referred herein as communities). Kebeles are the smallest administrative units in the government structure and may contain about 500 households each. Two of the three communities (Guba-Sherero and Holagoba-Kukie) were selected from rural Halaba, a Woreda (~District) in Southern Nations, Nationalities and People’s Region (SNNPR). The Woreda/District is located ~85 kilometers northwest of Hawassa, the capital of SNNPR. It is known for growing pepper and pulses, which are considered cash crops for the farmers. The third rural community (Edo-Qontola) was selected from Adami-Tulu-Jido-Kombolcha (ATJK) District near Zeway town, in the Regional State of Oromiya. It is located about 160 kilometers southeast of Addis Ababa, Ethiopia’s capital. Maize, teff, wheat, barley and different oil seeds are the major crops produced in the district. The area is characterized as dry land with both irrigated and rain-fed crop productions. The study communities were part of a larger Ethiopia-Canada project between Hawassa University (Ethiopia) and the University of Saskatchewan (Canada) that sought to improve agricultural productivity and human health in south Ethiopia [20]. In all the three communities, the study population were mothers and their <5yrs of age children. Hence, the inclusion criteria were households in the community with mothers and children <5yrs of age. The sample size was determined using formula for cross-sectional studies in each community [21]: no=[Z1−α22p(1−p)]/d2 and n = noN/[no + (N − 1)] where no, n stand for sample sizes before and after applying “finite population correction” factor, respectively; Z = 1.96, p = probability of expected prevalence, N = total population of interest and d = margin of error. In this study, prevalence of maternal undernutrition was taken as expected prevalence; p = 0.27 with a margin of error (d) of 5% for calculation of the sample size. The calculation yielded 200 households with mother-child pairs per each community. Adding a 5% contingency, a total of 630 mother-child pairs, (i.e. 413 from the two communities in Halaba and 217 from the third community in Zeway) were required. Sample size in Halaba was higher as two of the selected communities from this district were merged for this analysis. The selection of additional community in Halaba was needed for a subsequent intervention-control study. Selection of individual households was carried out by first obtaining a list of all eligible households (those with mothers and under five children) from the local health post and applying simple random sampling. In this study, a household was defined as one that had a mother and at least one under five years of age child and was served by the local health-post regardless of whether polygamy was practiced or not. Whenever households had more than one eligible child, the youngest was considered. The study was approved by the University of Saskatchewan Behavioral Ethics Board (BEH #12–357) as well as the Regional Health Bureaus of SNNPR and Oromiya. All mothers gave oral consent to participate in the study. Due to the low literacy rate in rural Ethiopia, obtaining written consent was not feasible. Oral consent was the most and culturally appropriate way of obtaining informed consent. The consent information was written in simple and easy-to-understand manner which was also translated into the local language. Consent forms were attached on a coded interview questionnaire for each participant. Female data collectors read and explained to each participant the purpose of the research as outlined in the consent form. Once participants gave oral consent to participate in the study, the data collectors wrote the participants’ full name on the consent form to indicate consent. This was then signed by the Principal Investigator (GE), detached from the interview questionnaire and stored in safe locker. The obtaining of oral consent was approved by the ethics committees. The study was carried out in March-June 2013. Information on the characteristics of participating households was collected by a questionnaire adapted from previous national and local surveys in the region [22–24]. We collected information on household size (number of usual members of the household), number of children <5yrs of age, marital status, polygamy, education of mothers and their husbands (no formal education = never attended school, primary = grade 1–6; post-primary = above grade 6), household headship, usual occupation, ownership of domestic animals, cultivable land, ownership of various household assets, housing characteristics (roof, floor, window material, number of rooms), sanitation facility, access to drinking water, persons responsible for fetching water and time required to fetch water, women’s access to their own farm land, person in charge of agricultural produce and some other variables. The questionnaire also contained a section that assessed access and utilization of health/nutrition services as well as dietary habits of mothers during their most recent pregnancy and lactation; these items were adapted from the baseline survey for national nutrition program [22]. Information on access and utilization of health/nutrition services included whether mothers had visited any healthcare facility, type of healthcare facility visited, number of visits, place of delivery of the youngest child, and whether mothers had received iron/folate supplement, as well as if mothers had received any health/nutrition education during/after their most recent pregnancy. Additional questions asked include whether mothers abstained from eating certain foods during pregnancy for cultural reasons including the type of foods avoided this way (food taboos), what their eating pattern was like (same, less, more than usual) during pregnancy and lactation, as well as whether mothers knew what balanced diet was. From the data collected, we created a set of variables to reflect gender, economic and demographic structures of the study households. Difference in years of formal schooling between husband and wife (i.e. years of formal schooling of husband minus that of wife) was used as a proxy measure for empowerment index for the household. The use of years of formal schooling to measure empowerment is supported in literature [25, 26]. We also used the length of time required to fetch drinking water as a proxy measure to estimate work burden on women. WHO/UNICEF recommend that drinking water should be accessible within 30 minutes of round trip from one’s residence [27]; thus, if time required to fetch water took 30 minutes or longer, this would significantly increase work burden. Women’s access to own-piece of farm land, control over agricultural produce, and polygamy were also variables used to reflect the gender dimensions of participating households. Likewise, we used information on cultivable land size per household and number of persons per household to calculate physiological density (i.e. the number of persons in the household/unit of cultivable land) and which provided information on how much land was available to produce food for the family. We then placed participating households into two categories (≤8 persons/ha or >8 persons/ha) and explored associations with nutritional status of mothers and children. Ownership of livestock is another important resource for agricultural communities in Ethiopia. We used the number and type of livestock information to calculate Tropical Livestock Units (TLU) for each household and divided the households as having low, average or high TLU. One TLU is estimated as the equivalent of 250 kg livestock [28]. A wealth index (WI) was also developed for each participating household to classify households based on socioeconomic status. To achieve this, we used various assets owned by households and other housing and sanitation related characteristics. The assets include ownership of radio, TV, mobile phone, bicycle, horse/donkey cart, motorcycle, handheld torch and oxen. Housing characteristics include roofing structure (corrugated iron or thatched grass roof), flooring materials (cow dung smeared/cement or mud/earth), presence or absence of windows, crowding (persons per sleeping room ≤5 or ≥6) as well as presence or absence of an improved sanitation facility. The use of asset-based approach to determine households’ socioeconomic status is also common in DHS surveys at national level [1, 29]. It is usually used for poor countries where large proportion of the population does not have regular income. Various methods can be used to weigh each item (in our case all binary variables) and calculate the actual index [30]. Each household received a score of 1 or 0 depending on whether it owned a particular asset. We then weighted each binary variable by the inverse of the proportion of households that owned the particular asset or had the particular characteristics [30]. This method assumes that if assets are owned by just the few, it is an indication that those “few” are wealthier than those that do not own the asset, hence they are given greater weight. After calculating WI for each household, households were grouped into low, medium and high WI categories. Trained female data collectors (mainly nurses) who fluently spoke the local languages administered the interview questionnaire at participants’ residence. Data collection was supervised by the principal investigator (GE) and B.Sc. nutrition graduates who also spoke the local language fluently. Anthropometric measurements of mothers and their children were carried out at the nearest health facility (health post and health centre), local school campus or outside the local Kebele office. On separate dates, participants living near to any one of these locations were invited to attend the anthropometric measurement sessions. Those who could not attend were visited at their residence. One person (GE) conducted all anthropometric measurements for both mothers and children to avoid inter-measurer errors. Child anthropometry included weight (measured to the nearest 10g) using electronic scale (Seca 770, Seca Corporation, Hanover, MD, USA); height (for children ≥24 months of age)/recumbent length (for children <24 months) (measured to the nearest 0.1cm) using adult/infant length/stature measuring board (Perspective Enterprises, Portage, MI, USA); head circumference (measured to the nearest 0.1cm) using a flexible non-stretch tape; mid-upper arm circumference (MUAC) (measured to the nearest 0.1cm) using colored insertion tape for children; and triceps skinfold thickness (measured to the nearest 0.2mm) using skinfold caliper (Holtain Ltd, Crymych, United Kingdom). Mothers’ anthropometry included weight (to the nearest 0.1kg), height (to the nearest 0.1cm) and MUAC (to the nearest 0.1cm). All measurements were taken in duplicate and averages were considered when the duplicates were similar. If the values were not similar, a third measure was taken to obtain the average of the two similar values. Standardized procedures were employed when taking body measurements [3, 23]. Birthdates for children were determined from immunization cards while age of mothers was asked verbally. Questionnaires were inspected daily and errors or inconsistencies were corrected at the field level. Information from questionnaire was entered in SPSS computer package (IBM SPSS Statistics version 20, IBM Corp., Armonk, NY, USA) and cleaned by running simple frequency distributions. Univariate and bivariate analysis was performed for the descriptive statistics. Bivariate and multiple variable regression analysis were performed to explore associations of gender and socioeconomic-demographic variables with maternal and child undernutrition [i.e. Body Mass Index (BMI) < 18.5 kg/m2 and Length- or Height-for-age Z score (LAZ or HAZ) < -2 Standard Deviation (SD), respectively] using Chi square tests and Multiple Classification Analysis (MCA). Only variables that were significant in the bivariate analysis were included in the MCA. Maternal BMI and child LAZ/HAZ as continuous variables served as main outcome variables and results were presented with the associated Eta (η) and Beta (β) values, indicating the bivariate and multiple variable coefficients of variation, respectively. WHO Anthro (ver. 3.2.2) 2011 was used to analyse all anthropometric data for children. Mean length- or height-for-age z-score (LAZ/HAZ), weight-for-length/height z-score (WLZ/WHZ), weight-for-age z-score (WAZ), MUAC-for-age z-score, head circumference-for-age z-score, triceps skinfold thickness-for-age z-score (only for children ≥ 3 months) and BMI-for-age z-score were calculated. Differences were tested using t-test for independent samples. Prevalence of stunting (LAZ/HAZ <-2SD), wasting (WLZ/WHZ <-2SD) and underweight (WAZ<-2SD) were also calculated. Body measurements from mothers were directly entered in SPSS spreadsheet and average MUAC, weight, height, as well as maternal short stature (as %<145cm in height) were calculated [3]. BMI of non-pregnant mothers and severity of maternal undernutrition were estimated per WHO classification of BMI (i.e. <18.5 kg/m2 = underweight, 18.5–24.9 kg/m2 = normal and >25 kg/m2 = overweight/obese). Since MUAC is relatively stable during pregnancy [31], all mothers in the study (including pregnant ones) were grouped as undernourished or normal using MUAC <23 cm as a cut-off point. For comparison purposes and where possible, results were presented along with findings from national/regional studies. Statistical significance was set at a p-value of <0.05.