Study design and setting
A community based quantitative cross-sectional study design was conducted from Jan 08, 2012 to Jun 16, 2012 in Ahferom district located around 200 Kilometers from Mekelle, the capital city of the Tigray regional state. At the time of the study there were about 44,194 households and 207,712 residents in the district. Tigray region is among the 9 regional states of Ethiopia.
Eligibility criteria
Respondents who were 18 years and above and who had volunteered to interview were included in the study. Respondents who resided in the area for less than 6 months and those unable to communicate properly were excluded from the study.
Sample size
The sample size of this study was calculated from EDHS 2005 household iodized salt coverage in Tigray region [19] with the assumption that it was close to represent households in Ahferom district. Taking 5% margin of error and 95% confidence interval of certainty (alpha =0.05), the actual sample size for the study was computed using one- sample population proportion formula as indicated below. Ten percent of the sample was added considering of the non respondents. N = 44,194 households.
\( n=\frac{{\left({z}_{\alpha /2}\right)}^2\times pq}{d^2} \) Where, n = Sample size, zα/2 = Critical value 95% confidence interval = 1.96, P = proportion of being using iodized household salt = 0.25, q = proportion of not being used iodized household salt = 0.75, d (marginal error) = 0.05
$$ n=\frac{(1.96)^2\times 0.25\times 0.75}{(0.05)^2} $$
$$ n=288.12\approx 289 $$
Including 10% of non responsiveness, the final sample size was 318 households.
Sampling technique
A multistage with three stage sampling process was used to ensure representative of all residents in the district. In stage 1, to confirm uniformity kebelles (smallest administrative structure) were stratified by residence type. Kebelles were considered as uniform in characteristics, hence they were considered as clusters. In stages 2, 2/6 kebelles from urban and 6/26 kebelles from rural were randomly selected by the lottery method after listing all kebelles in both urban and rural areas separately. Probability proportion to the size of the households was used to allocate number of sampled households in each of the selected Kebelles for the study. In the final stage, systematic random samplings were used to select the households from each selected Kebelles. To do that sampling interval [K] was calculated to each selected kebelles. The survey in each kebelle was started after pinning pen to indicate the direction from where to start and the first household was selected after counting households of K/2 for even interval or K/2 + 1 for the odd interval from the first household contact. And the next households were selected every K + K/2 from the first selected household to the next household until it reaches to the allocated final sample size of each kebelles (Fig. 1).
Data collection tools and procedure
Data was collected using structured questionnaires which sought information on socio-demographic and economic variables, availability and accessibility of iodized salt, practice of salt utilization, concentration of iodized salt, Knowledge and attitude regarding to iodized salt and IDDs. The questionnaires were adapted from different studies taking into account the local situation of the study area [10, 21] (Additional file 1). The collected salt samples were tested by using an iodometric titration technique to measure the iodine concentration. This process was done in Tigray Health Research Laboratory.
The salt samples were collected from the top, middle, and bottom of the pack (bag) using a moisture free, clean plastic container with cover. The samples were labeled with the following information during collection: Date of sampling, name of the Kebelle, and house number. During the visits, we first explained the aims of the study to household food caterers. After obtaining informed consent, each participant was interviewed by trained data collectors (diploma nurses). In addition, three public health officers were recruited as supervisors. After completion of the interview to each household food caterers, they were requested to provide three teaspoons (15 g) of consumption salt for iodine concentration test.
Operational definitions
Adequately iodized Salt: It is a salt that is fortified with the iodine which is > = 15PPM.
Improper utilization: practicing at least one practice that reduces the iodine content or < 15 PPM Iodine concentration in the salt.
Food caterer: Household member responsible for cooking in most of the time.
Self-report of “Yes, I used iodized salt”: a food caterer who knows and used iodized salt within 24 h.
Data quality assurance
To ensure data quality and consistency of the measurement tool, the questionnaire that was in Tigrigna was translated back to English. About 5% of the total participants of the study were pre-tested by similar households to check any discrepancy. Data was collected under close supervision and data was checked for completeness daily by the principal investigators. The quality of the test of the iodometric titration technique was checked to positive and negative controls.
Data management and analysis
Data was entered to SPSS Version 16.0 and exported to SAS version 9.2 for analysis. Knowledge was assessed by asking a range of questions about IDD, iodized salt and marking the correct answers of subjects out of a hundred. Average knowledge scores 50% or less was labeled as “poor knowledge” [10]. Using a Likert scale attitudes was assessed with five possible responses. The responses was labeled “favorable” or “unfavorable” as follows; For positive statements, responses including strongly agree and agree were labeled as “favorable” and disagree, strongly disagree and uncertain were labeled as “unfavorable”. For negative statements, those who responded “strongly agree”, “Agree” and uncertain was labeled as “unfavorable” and strongly disagree, disagree was labeled as “favorable” response. If the average attitude scores were greater than 50% it was considered as a favorable attitude [11].
Descriptive statistics were done to determine the proportion of households using adequately iodized salt, socio demographics and concentration of iodine in the salt. An inter observer variation of the iodized salt between self report of the respondents and iodometric titration was measured by using kappa statistics. The Kappa agreement was interpreted according to the scale [22]. Specificity, sensitivity, positive predictivity, negative predictivity and predictive validity of self-report on the use of iodized salt were calculated to check its validity with iodometric titration.
The outcome variable was a dichotomous outcome (1 = proper iodized salt utilization and 0 = improper iodized salt utilization). The analytic approach to modeling this type of data was the logistic generalized estimating equation (GEE), which takes into account the correlated nature of the responses. The order of responses within a cluster was arbitrary; therefore it was considered exchangeable and independent correlation structures [23]. The specified probability distribution was binomial with logit link function and the working correlation matrix structure was exchangeable (with the small Quasi likelihood under Independent Criterion (QIC)). The covariance matrix was robust estimator, and the scale parameter was Person chi-square (χ2). The main effect was the term used to build the reported model, and Kernel was specified for the log quasi-likelihood function. Bivariate logistic regression was used to see the strength of associations and factors that was found significant at p-value<=0.05 at bivariate analysis was entered to multivariate logistic regression to specify the independent predictors. Odds ratios (OR) were calculated to determine the strength of associations of the independent variables with the outcome variable at 95% Confidence Interval (CI).
Interactions of variables were assessed at p-value <= 0.05 and confounding of variables were assessed by backward and forward elimination and any variable which had > 20% change of coefficient of the parameters between the reduced and full model was considered as confounder [23]. Similarly Collinearity was checked by Variance Inflation Factor (VIF) and If VIF was greater than 10 it was considered as collinear and removed from the model.