Data source
Data were obtained from the last six Turkey Demographic and Health Surveys (TDHSs) conducted by the Hacettepe University Institute of Population Studies in 1993, 1998, 2003, 2008, 2013 and 2018. All surveys are nationally representative household surveys, and a weighted, multi-stage and stratified cluster sampling was performed on all surveys and to a large extent similar questionnaires were used [11, 12, 14,15,16,17]. These similarities made it possible to pool data sets and to obtain a large data set. Face-to-face structured interviews were conducted with women 15 to 49 years of age.
Data for children who were alive at the time of the TDHS fieldwork and who were born in the 3 years preceding the survey were included in the study. To obtain vaccination data for each eligible child, mothers were asked whether they had a vaccination card for the child, and if so, to show the card to the interviewer. The vaccination dates and doses were copied from the card to the questionnaire. If a vaccination card was not available for the child, then the mother was asked a series of questions in order to determine the vaccination status of the child.
We restricted the study sample to children aged 12–35 months and living with their mothers during the survey. If a mother had more than one child in this age range, we selected the younger child for inclusion in our analysis. A child was not included in the analysis when no information on vaccination status was provided (missing cases), but if the mother did not remember whether the child had ever been vaccinated, the child was treated as a zero-dose child. Finally, a total of 8198 mother–child pairs were eligible and included.
Variables
The dependent variable in the study is vaccination status, which is a binary variable with a value of “1” if the child is zero-dose and a value of “0” if the child is vaccinated. In addition to the survey year variable, which measures the period effect, variables related to household characteristics, parental characteristics, cultural characteristics and bio-demographic/health related characteristics were included in the study as independent variables. Factors related to household characteristics included environmental factors such as place of residence (urban/rural), region and household wealth. Parental characteristics consisted of the mother’s level of education, the father’s level of education (No education/primary incomplete, primary or secondary and higher), parental working status (at least one having social security, neither having social security) and health insurance. Cultural variables included bride payment, arranged marriage and mother’s native language (Turkish, Kurdish but also speaking Turkish, Kurdish, not speaking Turkish or other). Bio-demographic and health-related factors such as the mother’s age at birth (< 20, 20–34, or ≥ 35 years), the sex of the child, the age of the child, parity and birth interval (1st child, 2nd child, interval < 24 months; 2nd child, interval ≥ 24 months; ≥ 3rd child, interval < 24 months; 3rd child, interval ≥ 24 months), number of antenatal care visits (none, 1–3 or ≥ 4), place of delivery (home or healthcare facility), and previous tetanus vaccination of the mother (yes or no). Variables such as the father’s level of education, parental health insurance, and arranged marriage were not included in multivariate analyses so as not to cause multicollinearity. Previous tetanus vaccination of the mother is only included in the descriptive analysis as the data is present only in the 1993 and 1998 TDHS surveys.
Sampling designs
While the sample designs and questionnaires of the six surveys were the same, sample sizes differed. Hence, the number of women interviewed and the number of children whose vaccination information was collected were different. In order to avoid possible biases in the analyses stemming from the different number of observations in different surveys, weighting factors obtained with the Eq. 1
$${1\div({\text{a}} \times {\text{n}}_{{\text{c}}} /{\text{n}}_{{\text{T}}})}$$
(1)
were used [21], where a is the number of surveys, nc is the number of respondents for survey c, and nT is the total number of respondents for all surveys.
Statistical analysis
Analyses were carried out using the IBM SPSS 23.0 complex samples module, taking into consideration the sample design of the survey, which was selected through a multi-stage, stratified cluster-sampling approach, and taking into account the simple non-random sample selection of the data set. In the complex samples procedure cluster, strata and weight variables in the related TDHS surveys data were all accounted for. We calculated zero-dose prevalence and its 95% confidence interval (95% CI) for each sub-population group taking into account the complex survey design. Binary logistic regression analysis was used as the technique to identify the factors associated with zero-dose children. CSLOGISTICS command was used to perform complex samples logistic regression. In the binary logistic regression, the Exp (B) (odds ratio) values obtained from the analysis indicate the probability of the failure of the dependent variable in relation to the likelihood of its fulfillment. The Exp (B) value of an independent variable shows in which direction and to what extent it affects the likelihood of the fulfillment of the dependent variable. Values of less than 1 indicate that it reduces the likelihood of fulfillment, whereas values greater than 1 indicate that it increases the likelihood. Descriptive, univariate, and multivariate analyzes were conducted to measure the impact of independent variables on zero-dose children. The analyses were made at three levels: Turkey overall (national), the East region and other regions (West, South, Central and North regions). As the differentiation is predominantly between the East region and other regions and the number of observations in the West, South, Central and North regions is not sufficient, these four regions were evaluated and analyzed together.
Ethics
This study was a secondary data analysis of DHS data, which was approved by the institutional ethical review board of Hacettepe University, Turkey. All respondents undergo an informed consent process for participation in the surveys. Additional ethical approval was not needed for this study as it used publicly available data from from the Institute of Population Studies.