We used data from the 2016 Nepal Demographic and Health Survey (NDHS), which is a national level household survey. Under the Ministry of Health, the survey began in June 2016 and lasted until January 2017. A total of 11,473 households were selected for the sample, where, 11,203 were occupied. From those households, 11,040 were successfully interviewed. Among the interviewed households, 13,089 women age 15–49 were identified for individual interviews; interviews were completed with 12,862 women, with a response rate of 98%. We restricted our analyses to individual women age 15–49 of reproductive age who given birth 5 years before NDHS 2016. Women of age group 15–49 selected for the study either were permanent residents of the household selected for an interview or an eligible visitor who had stayed there a night before the survey .
The data collection for the NDHS, 2016, was ethically approved by the Nepal Health Research Council (NHRC), which is a national level government organization leading in research activities in Nepal. Similarly, ethical clearance was also obtained from ICF Macro Institutional Review Board, Maryland, USA. Data were collected after an online application was submitted to the demographic health survey (DHS) program explaining the purpose of the study, intended use, and people who would have access over the data . After the review of the online application, permission had been obtained from the monitoring and evaluation body of DHS globally, MEASURE DHS, to use the data set for this study. The NDHS, 2016 data are publicly available at the USAID DHS program (at http://dhsprogram.com/data) in different formats.
Our outcome variable was the attendance of at least four or more times ANC (ANC 4+) which is the recommended number of ANC visits by WHO.
Altogether 15 variables relevant to the study were selected and divided into contextual factors and individual factors [22, 23]. Nepal has recently adopted the federal system of government, whereby the formal names of the provinces are yet to be decided, hence, they are still known by numbers through 1 to 7. Province of residence, place of residence (urban/rural), age of the women, ethnicity (Brahmin/Chettri, Janjati, Dalit, Muslim and others), religion (Hindu, non-Hindu), women’s education (no education, primary, secondary and higher), partner’s education (no education, primary, secondary and higher), respondent’s occupation (didn’t work, skilled worker, unskilled worker, and agriculture), household wealth index (poor, middle class, rich), provincial GDP, sex of household head (male, female), health care decision-maker (women herself, herself along with someone else, solely others), exposure to the newspaper (not at all, less than once a week, at least once a week), exposure to the radio (not at all, less than once a week, at least once a week) and exposure to television (not at all, less than once a week, at least once a week).
A total of 4006 ever-married women between 15 and 49 years were eligible to be included in our study. We estimated the latest pregnancy and live birth in recent 5 years with the utilization of ANC service and the socio-economic differentials in these indicators by age, work, residence, religion, wealth, media exposure were analyzed using a multilevel logistic regression model.
The variables were divided as per the need for predisposing and enabling factors. After that, descriptive statistics were performed to find the frequency of various independent variables. A Chi-square test was executed to find the relation between the dependent variables (ANC4+) and the outcome variable (live birth). Furthermore, the variables were divided into contextual factors and individual factors for further analyses. Province, provincial GDP, household wealth index, religion, and ethnicity were considered as the contextual factors while the rest were the individual factors. Since GDP was not available from NDHS, the GDP of 2011 was taken from a data source that contains data related to Nepal . The multilevel nested structure of analysis comprised 4006 individuals (level 1) grouped into 380 primary sampling units (PSUs), which were wards in rural areas and enumerator areas in the urban area (level 2). Again, the PSUs were nested into the place of residence (urban and rural area) (level 3). Multilevel logistic regression was performed to test the association of contextual and individual independent variables with the number of 4+ANC visits in Nepalese pregnant women. Variables associated with 4+ANC at a significant level p < .05 were considered for the multivariable analysis. A three-level random intercept and fixed-slopes model structure with individuals nested within PSUs and PSUs within urban-rural cities were fitted to estimate the odds ratios (OR) and 95% CIs, indicating the likelihood of having a higher mean of 4+ANC visit. The overdispersion of the data was handled by using a three-level random intercept model.
Stepwise forward selection of variables in subsequent models was conducted to obtain a parsimonious final model for ANC visit, according to the theoretical framework (Fig. 1). The first and second models consisted of contextual predisposing and enabling factors, and the second and third models comprised of individual predisposing and enabling variables, respectively. Variables that remained statistically significant at 5% (P ≤ .05) were retained in the analysis for adjustment in the next model. Thus, the final models included all significant contextual and individual predisposing and enabling factors. The statistical analyses were executed using R version 3.6.3 with the “lme4” package.