2.1 Data source and study population
We used the first Myanmar Demographic and Health Survey (DHS) [13], a cross-sectional nationally-representative survey conducted by the Myanmar Ministry of Health & Sports (MOHS) between December 2015 and July 2016. The survey followed a stratified two-stage sample design based on the 2014 census, and was weighted to allow representative estimates for urban and rural areas as well as for each of the seven States and eight Regions of Myanmar. The MOHS interviewed 12,885 women age 15-49 years old from 12,500 households and responses from women who delivered between 2011 and 2016 were analyzed. For this study, we utilized the structured Woman’s Questionnaire developed for the global DHS program and modified for the Myanmar survey, which included questions on maternal demographic and socioeconomic characteristics as well as antenatal and delivery healthcare services.
2.2 Assessment of HepB-BD vaccination
HepB-BD vaccination was defined as receipt of the HepB-BD within 24 hours of delivery. Participants’ children were coded as vaccinated within 24 hours based on documentation in the child’s health card or maternal report. Children were coded as not vaccinated if the health card did not document the HepB-BD vaccine or participants denied receipt or if participants didn’t know their child’s vaccination status. In a post-hoc sensitivity analysis we limited the definition of vaccination to the children who had the vaccine documented on their health card (185 participants, 5%). The vaccination status of the most recent child was analyzed. For multiple gestation deliveries, the status of a single child was assessed, as all twins and triplets either had missing (<1%) or concordant data.
2.3 Conceptual model
Our evaluation of factors potentially associated with HepB-BD vaccination draws on two conceptual frameworks. Andersen’s model of medical care utilization emphasizes the complementary contributions of predisposing factors (such as demographics and social structure) and enabling factors (such as the availability of medical care) [14]; Nutting’s model of health system performance highlights the sequential nature of health service utilization [15]. A medical care utilization model is appropriate to conceptualize access to vaccines given at birth that unlike most other Expanded Program of Immunization (EPI) vaccines are administered by clinical staff who provide obstetric and neonatal services, and as such are closely correlated with other processes indicative of an accessible and well-functioning clinical care system. Similar to many low-income countries, in Myanmar the monovalent HBV vaccine is not administered by EPI outreach staff; although MOHS policy permits midwives to administer vaccines to newborns at home, the MOHS has not procured monovalent vaccine doses for this purpose [16].
A simplified two-step pathway can be used to describe HepB-BD vaccination in Myanmar: first, the mother must deliver in a clinical care system where the vaccine is available; second, that system must administer the vaccine within 24 hours after birth. Socioeconomic inequities in access to ANC and delivery care have been documented in Myanmar [17]; inequities in treatment are less well understood but are plausible due to possible out-of-pocket costs [18], discrimination in healthcare settings [10], and other reasons.
2.4 Predictor variables
Potential predictors were selected based on their prior associations with receipt of the HepB-BD [20-26] or other childhood vaccines in Myanmar [27] or their theoretical associations with receipt of the HepB-BD. Continuous variables were categorized a priori. Maternal and infant characteristics included maternal age at delivery, parity, maternal education, urban residence (vs rural), geographic areas categorized in four zones: Delta (Ayeyarwaddy, Bago, Yangon), Coastal (Mon, Rahkine, Taninthayi), Central (Magway, Mandalay, Naypyitaw, Sagaing), and Hilly (Chin, Kachin, Kayah, Kayin, Shan), household wealth index quintile (calculated by the DHS using principal component analysis of household assets, building materials, and sanitation) [28], and low infant birthweight (< 2,500g). Health system variables were organized into domains related to antenatal- (ANC), delivery- and postnatal care. Recommended ANC services was defined as the combination of 3 tests (blood pressure, urine, and blood) with care from a skilled provider (physician, nurse, midwife, or lady health visitor), and modeled as a binary variable versus less or no ANC. Prepartum tetanus vaccination was included as an complementary indicator of ANC quality. Among participants who received any ANC, early utilization of ANC (attending the first ANC contact during the first trimester), the number of ANC contacts, and ANC location (government hospital only, government clinic only, private facility only [private hospitals and clinics, Myanmar or international non-governmental organizations, or other locations], home only, or multiple locations) were also evaluated. Delivery care predictors included delivery by a skilled provider, delivery location and mode of delivery; postnatal care was indicated by receipt of a postnatal health check within 24 hours by a skilled provider. Additional predictors included year of delivery and weekly media exposure (use of either print, radio or TV media at least once per week, vs less often).
Survey responses of ‘don’t know’ (early utilization of ANC, number of ANC contacts, prepartum tetanus vaccination, birthweight, postnatal health check) and ‘unmeasured’ (birthweight), were combined with missing values and labeled as a separate category in the analysis accounting for 37% of predictor variables. The percent of missing data for each predictor was as follows: 2% of prepartum tetanus vaccination and mode of delivery, 5% of location of ANC, 6% of early ANC and number of ANC contacts, 24% of postnatal health check, 50% of low birthweight. In a sensitivity analysis we used multiple imputation with chained equations to create 50 datasets with imputed values for all variables, and no substantial differences in the estimated effect sizes were found when compared to the multivariable model without multiple imputation (data not shown).
2.5 Statistical analysis
Participant and health system characteristics were described with frequency distributions and differences between groups were evaluated with the Rao‐Scott χ2 test. The health concentration index [29] was calculated to evaluate inequality of vaccination rates ranked by household wealth and maternal education (described in Appendix 1). Two logistic regression models were used to assess factors associated with HepB-BD vaccination. First, a model was fit using predisposing and enabling factors (geographic region, delivery year, and maternal characteristics) to assess demographic and socioeconomic patterns of HepB-BD vaccination (Model 1). Second, we added to the model health system factors, as well as additional predictors that had significant (p ≤ 0.2) univariate associations with the outcome (parity, media exposure, Cesarean delivery, postnatal health check). We then used Allen-Cady modified backwards selection (Wald test, p < 0.1), resulting in the removal of parity and the postnatal health check from the final model (Model 2). Colinearity was evaluated using weighted linear regression, and the Variance Inflation Factor (VIF) was < 3 for all predictors, signifying low concern for collinearity, while a link test and the Archer–Lemeshow Goodness-of-Fit test [30] indicated no evidence for specification errors, missing interactions or nonlinear variables. Odds Ratios (OR) and their 95% Confidence Intervals (CI) were used to estimate the association of factors with the outcome; predicted probabilities were calculated using average marginal effects at representative values of maternal and health system characteristics. Data were analyzed using survey features that incorporated the complex sample design and survey weights. All p-values were two-sided and p < 0.05 was considered statistically significant. All analyses were performed with Stata 16.0 (StataCorp, College Station, TX, USA), using the conindex program to calculate concentration indices [31].
2.6 Ethical approval
The present study was considered exempt by the UCSF Institutional Review Board. The survey protocol was reviewed and approved by the Ethics Review Committee on Medical Research including Human Subjects in the Department of Medical Research of the Myanmar Ministry of Health and Sports.