The study Area
This study was conducted in Dawro Zone, southwest part of Ethiopia. The zone has five districts and one town administration with a total land area of 4,437 km2. It has three different agro-ecological zones; high land with altitude range of 2300 to 3200 meters above sea level (masl), middle land between 1500 to 2300 masl and low land below 1500 masl. The mean annual temperature and rainfall ranges from 15.1 – 27.50c and 1200 – 1800 mm respectively [24, 25].
The total population and health service coverage of the zone in 2016 was 702,517 and 93% respectively. Women of reproductive age group (15-49 years) accounted for 23% of the total population of the zone . IDD is endemic and iodized salt consumption is poor in the study area [9, 27].
Cluster randomized controlled trial was carried out for six months from January to June 2017. The unit of randomization was households served by health development armies (villages having 27-30 households). A health development army (HDA) in Ethiopia is defined as a network of women volunteers in the same village organized to promote health, prevent disease through community participation and empowerment. The HDA facilitate identification of local problems that hinder families from utilizing key maternal, neonatal and child health services and to come up with locally acceptable strategies to address the problem. They are also responsible frontline actors in discussing preventive health issues including health education and mobilize the community through different campaigns.
HDA inclusion and exclusion criteria
This trial was conducted in 24 randomly selected HDAs (clusters). The clusters were stratified in to highland and lowland by expecting knowledge, attitude and practice (KAP) differences in IDD and iodized salt utilization between the dwellers of these two locations. As indicated by previous studies, highland areas are more at risk of IDD [28, 29]. Therefore, people living in these areas have more exposure to visible sequale of IDD like goiter and could have better knowledge about it. All the selected clusters in highland and lowland kebeles with a minimum inter-cluster distance of 9kms (to minimize information contamination) were included in the trial. Middle land clusters located at the boarder of lowland and highland (having no clear boundary) were excluded.
Participant inclusion and exclusion criteria:
Individual participants were all women or reproductive age group (15-49 years) who were primarily responsible for meal preparation. Women who refused to give verbal consent, those who were mentally ill; those who had severe learning difficulties or planned to leave the area within next one year were excluded.
The study zone had five districts and one town administration. Two of the districts were selected by simple random sampling (SRS) and the town administration was included. The list of kebeles (the smallest administrative unit in the district) as a sampling frame was obtained from each district and the town administration bureau.
The kebeles were stratified in to highland and lowland. Accordingly, 29 highland and 31 lowland kebeles were identified, sequentially numbered and three of them randomly drawn from each stratum. In the six selected kebeles there were 55 HDAs (clusters). The clusters were again numbered and one cluster was randomly assigned to intervention or control arm in each stratum. In this way, 24 clusters (12 in each arm) were assigned by a supervisor (health officer with BSc degree) who was unaware of the study-arm assignments. Due to the nature of intervention, blinding of the participants was not applied. The schematic presentation of the sampling procedure was summarized in figure 1.
The intervention was targeted at the cluster level. Clusters were health development armies containing 27-30 households in the village. From each household a women of reproductive age group who was primarily responsible for household meal preparation was selected to participate in the trial.
To execute the trial, first educational and training tools were prepared. The tools contain clear message on iodine, its source, benefits of iodized salt, how to handle and use iodized salt, how to store it and the consequences of not using iodized salt. To prepare these tools, we referred available resources from World Health Organization (WHO), United Nations children’s fund (UNICEF), Iodine Global Network (IGN), Ethiopian Demographic and Health Survey (EDHS), National micro-nutrient guide lines and adapted to local context and target groups [23, 30-34].
Health Development Army members selection for training
Two HDA members who fulfill the criteria (Volunteer, 10th grade completed, fluent in speaking local language, active participant in the group, stable and live in the area for the next one year) were selected from each intervention site. Five days training supported by demonstration and posters was given by principal investigator (PI). A copy of the teaching material was also given for each TOT (training of trainer) members.
Nutrition education for intervention groups
For intervention groups, a six month iodine nutrition educatio was given by using the prepared materials. Members of HDA who actively completed five days training handled the sessions. The schedule for the session was twice every month for six consecutive months. In each one hour
education session, two-way process of sharing message, active involvement of the participants, freely exchange of ideas and demonstration of iodized salt was practiced.
On the contrary to the intervention clusters, the control clusters did not participate in any nutrition education session. They continued with their routine health care services of the ministry of health (MOH) as usual. The MOH routine health care package for the frontline health workers in Ethiopia includes 16 components. These components can be categorized in to four major areas: family health (family planning, maternal and child health, nutrition and vaccination services), disease prevention and control (HIV/AIDS and STI, tuberculosis, malaria and first aid cares), hygiene and sanitation (promotion of sanitary latrines, waste disposal management, water supply, food hygiene and safety, control of insects and rodents, personal hygiene and healthy home environment services) and health education and communication .
Supervision, endline data collection and measurement
The sessions were monitored by PI and/or assigned supervisor every month. At the end of each three month, there was review meeting in which all members including the supervisor present their work in the presence of PI. The purpose of this three months review meeting was for monitoring and process evaluation to ensure the sessions were being implemented as planned or need any adjustment. At the end of intervention period (after six months), data were collected from both groups (intervention and control) using the same data collectors and questionnaire that was used at the baseline.
All factors and tools used were made similar for both arms except the intervention. This ensured that the observed change in intervention arm was reasonably attributed to the intervention. The change in KAP (increment in knowledge, change in attitudes or reduction of risky behaviors, and improvement of practice) was measured by comparing the baseline data with end line findings.
Deviations from the protocol
Generally, the intervention was carried out as planned. However, there were some deviations from the protocol. Initially, the intervention was planned for nine months, but we believe this is too long to see the effect of intervention and reduced to six months. As this is one of the unfunded studies, we could not be able to measure urinary iodine concentration of participants and failed to assess nutritional status as a secondary outcome. The last deviation from the study protocol was that the study lagged seven months behind the schedule because of the political instability in the country that interfered data collection.
The outcomes were change in mean score of the three composite variables (knowledge, attitude and practice) about IDD and iodized salt after intervention. The secondary outcome, according to our plan, was change in iodine nutrition status of participants after intervention. Unfortunately, this was not done due to resource limitations.
Sample Size Determination
The sample size was calculated using GPOWER 3:0 Software, by taking into account the intracluster correlation coefficient (ICC), the average cluster size, the expected effect size, and the power of the study. The ICC was assumed to be 0.04 from previous studies [36, 37], total number of clusters 24 with a minimum of 27 participants in each cluster and the expected effect size of 0.29. The effect size was estimated from the previous studies conducted in south and northwest Ethiopia [38, 39]. The study in southwest Ethiopia reported about 44.7% respondents had good knowledge and 42.6% had positive attitude towards consumption of iodized salt . Similarly the northwest Ethiopian study reported about half of the households’ (49.8%) added salt at the beginning and middle of food preparation (inappropriate practice). This means, the rest 50.2% added at the end of cooking process (appropriate practice) .
Therefore, most likely the current study zone might have similar proportions and we planned to increase good knowledge from 44.7% to 74%, positive attitude from 42.6% to 72% and appropriate practice from 50.2% to 79%. With these assumptions a power of 80% was anticipated to detect a difference in KAP scores between the two groups with α err prob. of 0.05. Finally, 10% expected non-response rate/drop-out was added to get 652 as the sample size for the trial.
Training was given for twelve data collectors (health extension workers) on how to use the questionnaire, ethical issue, revisiting closed houses and their work relationship with the supervisor for three days. The questionnaire was pretested on non-selected but similar adjacent kebeles. A week after pretest, baseline data were collected from both intervention and control clusters by face to face interview using structured and modified questionnaire. The same tool and data collectors were used to collect endline data after six months of intervention.
For face validity, 32 households (5% of the sample) were interviewed and each questionnaire was completed by data collectors. The completed questionnaire was collected and evaluated by the principal investigator, co-investigators and a statistician. For content validity, an expert of three specialists in endocrinology, nutrition and biostatistics examined the initial questionnaire. The experts were asked to comment on individual items in relation to the accuracy and content. Finally, items were modified slightly based on the findings from pretest and the expert reviews.
In addition, Cronbach’s α was determined for internal consistency (Cronbach’s α=0.93, 0.77 and 0.71 for knowledge, attitude and practice respectively). Addition of iodized salt at the end of cooking process, not washing the salt before use and not exposing it to sunlight are some of the recommended practices of iodized salt. Convergent validity was checked by observing the practice of women against these recommended practices during demonstration sessions.
Data were collected using a series of 16 questions about knowledge of IDD and iodized salt and eight questions about practice of iodized salt. In each case a correct response was given a score of one, whereas an incorrect response was given a score of zero. Then, knowledge and practice indices were produced by adding individual answers across items.
Similarly, data on attitude about iodized salt and goiter were collected using a series of seven questions. Each question had three options (yes, no or don’t know) and a three-point Likert scale (agree, disagree or neutral). The responses were dichotomized in to “yes” or “no” and “agree” or “disagree” as there was no response on third options (“don’t know” and “neutral”). A score of one was given for responses that defined positive attitude and a score of zero was given for negative ones. An index for attitude was produced after summing individual answers across items.
Finally, for all the three indices (knowledge, attitude and practice scores), we computed the difference by subtracting the baseline value from the endline. These differences were used as response variables for the subsequent analysis.
Data Processing and Analysis
Data were doubly entered into Epi Info version 3.5.4, cleaned and analyzed using SPSS version 21 (SPSS Inc. Chicago, USA). Descriptive statistics were summarized using frequency and proportions. The mean difference in KAP scores between intervention and control groups was tested using independent sample t-test. Principal component analysis (PCA) was used to construct a wealth index from 20 household fixed asset such as presence of latrine, source of drinking water, possession of television, radio, mobile telephone, availability of separate kitchen from living house, domestic animal and land possessed in hectare. These variables were dichotomized and coded ‘1’ for the household possessing the asset and ‘0’ for the rest. Finally, the factor scores ranked ordered into three relative measures of socio-economic classes (poor, medium and rich).
For this study, the primary outcome measures were composite variables (knowledge, attitude and practice scores) determined by summing all responses across items and subtracting the baseline sum from the endline to get the difference. These differences in scores were taken as an outcome variable for each.
As mentioned above, there were two measurements on the same subject (baseline and endline). Since repeated observations within one subject are considered not independent of each other, the standard linear regression could not allow for the repeated measures and could underestimate the standard error. In this case, the Generalized Estimating Equations (GEE) was used to extend the generalized linear model to allow for analysis of repeated measurements. GEE helps to correct for the expected within-subject correlations . Therefore, after checking all assumptions including multicolinearity (maximum VIF=1.240), we fitted generalized linear models (GEE).
First, we run bivariate analysis for each outcome variable. Then, to control for confounding and adjust for within cluster correlation of measurements, all variables with a p-value ≤ 0.2 in the bivariate analysis were fitted into the model for multivariable analysis. All analysis was conducted according to intention-to treat principles and a p-value of 0.05 or less was considered statistically significant.