Study Setting
The study was conducted from February 2, 2016 to April 4, 2016 (the post-harvest, dry season) in the Fagita Lekoma woreda. Fagita Lekoma is one of 12 woredas in the Awi Zone, which is located in the Amhara Regional State. It is a rural woreda (20 of its 22 kebeles are rural) located 450 kms from Ethiopia’s capital, Addis Ababa [11]. We selected Fagita Lekoma because there are no SFPs in the woreda. The woreda has 6 health centers, only one of which provides outpatient care for severe acute malnutrition (SAM). The estimated population for the woreda is 156,671; with 36,435 households, and 21,213 children under the age of five. The five kebeles in this study had a total population of 22,682 [19].
Study design and population
This community-based prospective cohort study was conducted among children with MAM aged 0–59 months. Children 59 months of age or younger with MAM that lived in the randomly sampled kebeles were eligible for recruitment. We excluded children older than 59 months; whose age was not known; without MAM; with no present mother or whose mother was unable to communicate with the study team; children who had health problems or disabilities that made it difficult to collect anthropometric measurements; and children with MAM who were receiving medical treatment.
Sample size and sampling techniques
Because food security has been shown to be an important factor for predicting poor health outcomes in children with MAM we selected it as an “exposure” variable for sample size calculation and for the stratification of our Kaplan-Meier survival plots [12]. Households were categorized as food-secure and food-insecure based on Household Food Insecurity Access Scale (HFIAS) results that were from previous study [12]. We calculated our sample size using the double population proportion formula. Our assumptions were as follows: 37.78% children with MAM in food-secure households would have poor health outcomes [12] for an adjusted hazard ratio (AHR) of 1.39 for poor health outcome among food-secure compared to food insecure households [12]. We assumed a 95% two-sided confidence interval (CI), a statistical power of 80%, and a one-to-one allocation ratio of food-secure to food-insecure. Based on these assumptions, using EPI INFO 7 [20], we calculated a sample size 384. Allowing for an additional 5% non-response rate, the total sample size was 404 (202 for food-secure households and 202 for food-insecure households).
We randomly selected 5 kebeles from Fagita Lekoma’s 20 rural kebeles (25%) using a simple random sample lottery method. We then visited all households in the selected kebeles and screened all children aged 0–59 months (n = 2995) for their nutritional status. We used the conventional definition of MAM: having a weight-for-height (WFH) below the WHO median child growth standards (the child growth with Z-scores between -3SD to -2SD).
All children were assessed for WFH using WHO Anthro version 3.2.2 software and those with MAM were identified and registered. At this time we also categorized households as food insecure and food secure. We found 414 children with MAM (202 from food-secure and 212 from food insecure households). We retained all 202 children from food-secure households. We randomly selected 202 children from the remaining 212 food-insecure households using a lottery method. When there was more than one child with MAM in a household, we selected one of them using lottery method. The selected children were enrolled in the study and followed for two months.
Study variables and measurement
Our outcome variable was whether, by the two-month follow up visit, a child had progressed to severe acute malnutrition (SAM); had not recovered from MAM, or had died. Children with any of these outcomes were categorized as having “poor health outcomes”.
We categorize children as having MAM, if at the second follow up visit, they had a weight-for-height/length (WFH/L) between -3 and - 2 Z-scores (-3SD to -2SD of the WHO median value), or WFH/L at 70–80% of the National Center for Health Statistics (NCHS), or had a MUAC measurement that was > = 11.5 cm <12.5 cm, without edema. Children whose MAM status did not change by the 2-month follow up period were categorized as not recovering. We categorized children as having SAM if, at the first or second follow-up visit, they had WFH/L below −3 SD of the WHO median value and/or (WFH/L) below 70% of the NCHS median value and/or MUAC <11.5 cm, with or without edema. Children were categorized as recovering if, at first and/or second follow up visit, they had WFH/L Z-scores > = -2SD of the WHO median value and/or WFH/L > =80% of the NCHS, and/or MUAC > = 12.5 cm) with no edema.
Data collection methods
We collected data using a cross-sectional, structured, interviewer-administered questionnaire containing closed-ended questions and by taking anthropometric measurements of children and their mothers during home visits.
Our study began with the development of a project survey and the recruitment of project staff. Our survey was developed from standard, validated, English-language instruments that were translated to into Amharic. We recruited 2 health officers to supervise data collection, and 10 health extension workers and 3 nurses to act as data collectors. All spoke Amharic, the local language. We then conducted one-day training on how to collect the data for the data collectors and supervisors and then pre-tested the questionnaire in a kebele that was adjacent to our study kebeles, with 20 households (5% of our sample size).
The study had three data collection points: we collected baseline survey and anthropometric data during community-based nutritional screening for all children 0–59 months of age in our 5 sampled kebeles. We used the HFIAS to measure food security for stratifying the sample [21]. This tool is the current standard for assessing household-level food security and has been validated for use in Ethiopia [22]. Households that were enrolled in the study were visited once monthly for 2 months, during which mothers were asked follow up survey questions and anthropometric measurements of the study children were taken.
The survey contained questions on socio-economic factors, demographic risk factors, child characteristics, child-care practices, maternal characteristics, and environmental risk factors. We recorded the child’s vaccination status by reviewing immunization cards when these were available, or by using the mother’s recall. We checked bacille Calmette-Guerin vaccination by observing whether there was scar on the child’s arm.
We used procedures stipulated by the WHO to take anthropometric measurements [23]. Before measuring children we established their age, using a local event to establish the child’s birth period. Mothers were asked whether the child was born before or after certain major events until a fairly accurate age was pinpointed. If we were not able to determine the child’s age accurately, the next child in the household was recruited. We measured body length of children age up to 23 months (or those who were older but too ill to stand) in the recumbent position, without shoes, reading the length to the nearest 0.1 cm or 1 mm using a horizontal wooden length measuring board/sliding board. We measured MUAC for both the study children and their mothers. MUAC was measured on left mid upper arm half way between the olecranon process and acromion process using a non-stretchable strap, to the nearest 1 mm.
Quality control measures
We checked the calibration of the measurement scale by weighing a 2-kg stone after each child measurement and after moving the scale from one household to another. Then the scale indicators were checked against a zero reading before and after weighing every child and mother. Only one observer was used for each subject. Mothers and children were required to wear only light clothing in order not to skew the weight results. The project principal investigator reviewed collected data on a daily basis, and returned records with possible errors to the data collectors for correction.
Data processing and analysis
The collected data were checked for completeness, consistency and entered using EPI-data software; then the data were exported to SPSS version 20 for analysis [24]. Descriptive analysis such as Kaplan-Meier survival curves and log-rank test statistics were used to describe important variables of the study and compare the outcome variables. A Cox-regression model was fitted to identify risk factors for poor health outcomes of MAM. All predictors that were associated with the outcome variable in bivariate analysis at p-values of 0.20 or lower were included in our multivariate Cox-regression models. Crude and adjusted hazard risks with their corresponding 95% confidence intervals were computed. Variables with p-values <0.05 were considered statistically significant risk factors in this study.