Hepatitis B vaccination coverage across India: exploring the spatial heterogeneity and contextual determinants

The study is a cross sectional study and utilized the data from NFHS-4 which is publicly available through (https://​dhsprogram.​com/​data/​dataset/​India_​Standard-DHS_​2015.​cfm?​flag=​1). Thus, no further ethical approval is required. NFHS 4 was conducted during January 20, 2015 and December 4, 2016 across 640 districts spread over 36 states and union territories of India. Districts in India are the second basic and policy relevant administrative units. NFHS 4 is the Indian version of Demographic Health Survey (DHS) that used standard survey instruments across the country. NFHS used a stratified two stage cluster design to conduct the survey and used the 2011 census sampling frame to select the primary sampling units (PSU). Census enumeration blocks in urban areas and villages in rural areas constitute the sampling frame of PSUs. PSUs were then selected using probability proportional to size (PPS) from the sampling frame. Prior to the main survey, a complete household mapping and listing was done in the selected rural and urban PSUs and within the selected PSUs, the number of households (300 at least) were sub divided into segments of 100–150 households. And finally two of the segments were randomly selected using systematic sampling with probability proportional to segment size. In the second stage, 22 households were randomly selected with systematic sampling from the rural and urban clusters of segments. The details of sampling design, instrument and survey findings are available for public use [19]. The study sample consists of 199,889 children aged 12–59 months. Of the total 211,773 children, information on the three doses of HBV vaccine was not available for 11,884 children and these observations were dropped from the study.

During the survey, as a part of the core questionnaire, the mothers were asked to show the vaccination card to collect the information on various doses of vaccination including hepatitis B. In case mothers could not show or did not have vaccination card at the time of survey, they were asked whether the child received the doses of hepatitis B vaccine. A child is said to be vaccinated against hepatitis B if he/she was found vaccinated either in card or from mothers reporting. Those mothers who reported “Do not Know” were treated as not vaccinated (1.8% of total cases). This is the standard recommendation by Demographic Health Survey (DHS) to estimate the vaccination coverage among the children [20]. Among 199,889 children aged 12–59 months, only 109, 085 received all the three doses of hepatitis B vaccine. Although NFHS collects the information on the birth dose of hepatitis B, but NFHS provides an estimate of children receiving the three doses of hepatitis B received at 6, 10 and at 14 weeks from the day of birth, independent of the birth dose. As the present study is based upon NFHS data, we considered the last three doses of hepatitis B being received at 6, 10 and 14 weeks to create the outcome variable in this study. Table 6 in Appendix provides hepatitis B vaccine uptake information for all the three doses other than the birth dose among the study children.

It is likely to be some recall bias (non-sampling bias) in the data and it could be in either direction-over reporting or under reporting. Although, checking the validity of mother’s recall was beyond the scope of the study but to reduce the non sampling bias due to mother’s recall, we controlled the socio-economic and demographic factors which mostly determine the pattern of recall bias among mothers [21]. In another account to take care of the sampling bias, we used the “svy” command in Stata version 12.0 SE (STATA Corp LP, College Station,TX) with sampling weights to address the corresponding sampling bias and to get the unbiased estimates.

The analyses have been carried out at district level and at individual level (child). Prior to unit level analyses, the district level analysis is a comprehensive effort to understand the analogy of district level coverage of hepatitis B vaccination and its determinants in a spatial setting because, after states, district is the second administrative and policy relevant unit where demographic events and population health indicators are estimated to track and monitor the health conditions of the general population in India.

The outcome variable for the district level analysis is the proportion of children aged 12–59 months who received 3 doses of hepatitis B vaccine. In case of child level analyses, the outcome variable is the hepatitis B vaccination status of the child (whether immunized against hepatitis B or not). A child who was given all the three doses of hepatitis B vaccines considered to be vaccinated against hepatitis B virus otherwise not. Thus, the hepatitis B vaccination status of a particular child is a binary variable where ‘1’ is yes which denotes the child received all the three doses and ‘0’ otherwise.

A set of socio-economic and demographic indicators at the district level were used to predict the coverage of hepatitis B vaccine. These include the - (1) percentage of women with 10 or more years of schooling, (2) percentage of mothers who had full antenatal care (ANC), (3) percentage of mothers who received postnatal care (PNC) from a doctor/nurse/LHV/ANM/midwife/other health personal within 2 days of delivery (4) percentage of mothers who received financial assistance under Janani Suraksha Yojana (JSY) scheme for births delivered in an Institution, (5) percentage of institutional births (6) percentage of households with electricity connection, (7) percentage of households with an improved drinking-water source,² (8) percentage of children under age 5 years whose birth was registered and (9) percentage of breastfeeding children receiving an adequate diet.³

District level variations and determinants were examined using spatial analyses. According to Census of India 2011, there are 640 districts across 36 states and union territories with an average population size of 2 million [22]. These districts vary enormously in demographic, social, economic and health indicators. NFHS-4 for the first time had the distinction of providing demographic and health estimates at the district level.

To understand the spatial clustering of immunization across districts, Local Moran’s I indices were generated to measure the spatial autocorrelation. Similarly, bivariate LISA was used to analyze the association of certain characteristics of regions (districts) with the hepatitis B vaccine coverage across those districts. Such analyses has been increasingly used to understand the spatial heterogeneity in terms of demographic and public health indicators across the population [23]. Moreover, district level spatial analyses are helpful to assess the geographical disparity in health or other concerned indicators and identify the geographical pockets underprivileged in terms of the same [23, 24]. To check the empirical associations between the outcome and independent variables of the study, we preliminarily estimated the ordinary least square (OLS) model and conducted spatial diagnostics of the residuals in OLS model. As the event of study showed a statistically significant Moran’s I, we built up the spatial autoregressive models-spatial lag and spatial error model.

The India shape file was obtained from GitHub through https://​github.​com/​datameet/​maps/​tree/​master/​Districts and was used under the Creative Commons Attributions 2.5 India license. The projection of the map was in WGS 1984 UTM zone 43 N.

The set of independent variables used in the unit level analyses are child level characteristics, maternal characteristics and household characteristics. The child level characteristics include age of the child (in months), year of birth, sex, birth order, child lives with whom. Age of the child is categorized into four groups (12–23, 24–35, 36–47 & 48–59) while the year of births of the study children are 2010, 2011, 2012, 2013, 2014 & 2015. The birth order refers to the order of the child among all live births to a mother and labeled as first, second, third, fourth etc. Previous studies also explored the variation in child health care utilizations in terms of the birth order of the child [25]. Sex of the child is another important variable considered in this study to find the gender differential of hepatitis B vaccine coverage among the study group of children. Sex is a bio-demographic characteristic of the child and children are classified as male and female. To understand the care given to the children and health care utilization for the children it is important to know whether the child lives with their mother or not. Although we did not find any previous study but we assume that children living with a parent are likely to receive better care. Here the variable is categorized into two following categories- children living with mother & lives elsewhere. The maternal characteristics included in the analyses are mother’s educational attainment, caste and religion. Mother’s educational attainment is classified as no education, up to primary educated, completed secondary education and higher secondary or more educated. Caste is another important social variable that depicts the economic and social well being of the households in India. In India, the population is classified into four caste groups, namely, scheduled caste (SC), scheduled tribe (ST), other backward class (OBC) and others. Among these groups, SC & STs are the most under privileged and secluded groups in India. The national, state and local government in India provides reservation benefits to SC, ST and OBCs in education, employment, health and other related programs. Similarly, we have used four religion groups (Hindu, Muslim, Christianity and others) in the analyses and these three are the major religious groups in India. Wealth quintile in DHS data is a measure to capture the economic well being of the household. It is derived from a set of 37 asset based variables using principal component analysis (PCA). For analytical purpose, the wealth index is grouped into five categories-poorest, poor, middle, richer & richest. Besides, we have used place of residence (rural/urban) and region (North, South, Central, West, East and North East) in the analyses. These classifications are similar to that of NFHS-4 [26].

Bivariate analysis was used to understand the differentials in vaccination coverage by socio-demographic characteristics. The conditional probabilities were estimated for hepatitis B vaccination conditioned on the background characteristics of the children. And the multivariate logistic regression was used to understand the determinants of hepatitis B vaccination at individual level. Child’s hepatitis B immunization status (received all the three doses of the vaccine -yes/no) has been modeled and adjusted to a set of independent factors. A total of 1, 99,889 children aged 12–59 months consisted the unit level analysis of this study. Stata version 12.0 SE (STATA Corp LP, College Station,TX) was used to analyse the data.

Figure 1 gives the quintile map showing the distribution of hepatitis B vaccination across districts of India. A total of 121 districts had hepatitis B vaccination coverage of less than 41% (dark brown), 135 districts had between 41 and 51%, 154 districts between 52 and 62% and a total of 230 districts had more than 62% coverage. Geographical disparities and the gradual spatial progression from low to high can be observed in the coverage of hepatitis B across the districts. A high coverage of hepatitis B vaccine is highly clustered in southern and south-eastern parts of India. Very low coverage of hepatitis B vaccine can be observed mainly in western, north-eastern regions and some parts of central India. Table 7 in Appendix provides the spatial dependence of hepatitis B vaccine coverage and each of the indicators across the districts of India. And the corresponding Moran’s I value indicates the neighborhood effect.

Figure 2 presents univariate LISA cluster and significance maps of hepatitis B vaccine coverage in districts of India. The Moran’s I statistic was 0.628, suggested a highly significant spatial dependence in the coverage of hepatitis B vaccination in India. A total of 106 districts from Gujarat, Rajasthan, some parts of Uttar Pradesh and North-Eastern states formed the cold spots (low-low coverage) while 111 districts from West Bengal, Punjab, Himachal Pradesh, Orissa, some parts of Haryana, Chhattisgarh, Andhra Pradesh, Maharashtra, Kerala, Tamil Nadu and Telangana formed the hot spots (high-high coverage) across India. A total of 13 districts have been found as spatial outliers (high-low or low-high) of hepatitis B vaccination coverage in the country.

Table 1 presents the bivariate spatial association of hepatitis B vaccine with the socio-economic and demographic indicators. District level percentage of full ANC, PNC, women with 10+ years of schooling, institutional delivery and birth registration showed a dominant and statistically significant association with the coverage of hepatitis B vaccination among the children across the districts of India. The value of Moran’s I was lower for ‘mother’s receiving financial assistance’, ‘safe drinking water’ and ‘breastfed child receiving adequate diet’. This indicated that utilization of maternal health care utilization among the mothers across districts is substantially associated with HBV vaccination coverage among the children.

Table 2 gives the estimated results from the ordinary least square (OLS), SLM and SEM model for hepatitis B vaccine in the districts of India. OLS model gives the adjusted estimates without considering the spatial correlation into account whereas the spatial auto regressive (SAR) models give the association between the predictors and hepatitis B vaccine coverage by considering the spatial effects into account. In all the three models, we found education of mother, utilization of full ANC, PNC and financial assistance to be the highly statistically significant predictor of HBV vaccine coverage across the districts of India. Based on the model diagnostics of the spatial models, SLM model gave the best fit to the data under study with the lowest AIC value. The SLM model showed that except ‘birth registration’, ‘drinking water’ and ‘breastfed child receiving adequate diet’ all the predictors have statistically significant association with hepatitis B vaccination coverage. In the SLM model, the coefficient of PNC (β = 0.33, p-value< 0.01) was the largest followed by full ANC (β = 0.15, p-value< 0.01), mothers received financial assistance (β = 0.12, p-value< 0.01) and education of the women (β = 0.12, p-value< 0.01). This indicated that post natal care utilization among mothers across districts is strongly associated with the rate of immunization (hepatitis B) among the children in India. Interestingly, in the first model (OLS model) proportion of breastfed child receiving adequate diet showed a significant association with the coverage of HBV vaccine but once the spatial autocorrelation was adjusted the effect became insignificant in the SLM as well as in the SEM model.

After a- district level meso scale examination of the hepatitis B vaccine coverage in the previous section, this section describes the coverage of hepatitis B vaccination at individual level and it’s socio-economic and demographic determinants among the children across India.

In the study sample we found that 48% of the children were female and 52% were male. Of the total children, 76% of the children were from rural areas whereas 24% were from urban areas. A significant portion (26.5%) of the children belongs to the poorest wealth quintile. Around 72% of children were Hindu, 8% were Christian and 16% were Muslim. About 20% belonged to the scheduled caste and 21% belonged to the scheduled tribe (ST) group. The study population of the children were found to be of different birth orders. Around 37% of the children were of the first birth order whereas 17% of them were found to be of fourth or higher orders. And the mean age of these children was 35.5 months.

Table 3 gives the coverage of HBV vaccine in India, Gujarat and Punjab by background characteristics. Punjab had the highest and Gujarat had the lowest coverage of hepatitis B vaccine. The vaccination coverage did not show large variation across rural and urban areas. The variation in hepatitis B vaccination was large by wealth quintile. Children from the poorest section showed the lowest coverage than the rest of the wealth groups with a distinct gradient of higher vaccination rate over the improved wealth status categories. For example, the coverage of hepatitis-B vaccine was 47% in poorest wealth quintile compared to 64% in richest wealth quintile and the pattern was similar in Gujarat and Punjab. Religion differential of HBV vaccine was also evident. In the state of Gujarat, we found that the coverage was highest among Christian children (58%) compared to Hindu and Muslim children. Overall, this state showed a comparatively lower vaccination rate among the different religious group of children than the national rate. It was observed that Muslim children carried the burden of lowest coverage of HBV vaccination among them.

Sex of the child did not show any substantial difference in the HBV vaccination whereas birth order of the child showed a differential. Children of first birth order showed the highest coverage of the vaccine than the older birth orders. Results suggested that with the increasing age of mothers, a drop had been observed in HBV vaccine coverage in India as well as in Gujarat but not in Punjab. The coverage of HBV vaccine was found to be higher among the children who lived with their mother than otherwise in India. Similarly, mother’s education showed considerable difference in the coverage of hepatitis B vaccine among children and the coverage was comparatively more among the children whose mothers were higher educated than their counterpart. Similarly, the education gradient held true for the two states as well. In Gujarat the HBV vaccination was notably low with only 21% coverage among the children of no educated mothers.

Table 4 gives the conditional probabilities of HBV vaccination among the children by background characteristics. The conditional probability that a child in urban area will receive HBV vaccine is 0.54 while that in the rural area is 0.49. The sex of the children shows the same chances of being HBV vaccinated. Wealth status showed a clear increasing probability over the gradients and the chance was highest (0.61) of being HBV vaccinated if the child belonged to the highest wealth quintile whereas the chance was lowest (0.42) among the poorest children. Of the two caste groups, children belonging to SC showed the higher chance (0.52) of being vaccinated than the ST children (0.45). The chance of vaccination was maximum (0.55) for the first birth order and eventually the chance was as low as 0.35 for the six and higher ordered births. The probability was observed maximum among the children aged 12–23 months followed by the older ages with the lowest chance of being vaccinated among the 48–59 months age group of children. It was found that the chance was 50% if the child lived with the respondent otherwise the chance was around 36%. Educational status of the mother also showed an increased chance of HBV vaccination among those children of higher educated mothers. Among all the children, the chance was lowest (0.41) among those whose mothers were not having any formal education.

Table 5 shows the estimated adjusted odds ratio values (AOR) from the logistic regression showing the empirical association between HBV coverage among the study children and its predictors. Age pattern of HBV vaccination showed that compared to the 12–23 aged children, children of older ages were less likely to receive the vaccine. As the data are a cross sectional data and the current age of the children varied between 12 to 59 months it could be said that children born earlier were less likely to receive the vaccine than those children who born later. This result is consistent with the bivariate result and reflects the recent awareness about the importance of HBV vaccination. Furthermore, it could be an indication of the increasing financial burden associated with the subsequent children over the period of time.

Birth order of the children showed a statistically significant association with the vaccination. And higher ordered births showed a lower likelihood of HBV vaccination compared to the first. Sex of the child did not show a statistically significant association with HBV vaccination status. We observed mother’s educational status to be highly significant for HBV vaccination. Results suggested that children of not educated mothers were less likely to be HBV vaccinated than their counter part. Religion of the children and HBV vaccination showed a statistically significant association and Christian children were more likely to receive HBV vaccine than the Hindu children whereas the Muslim children were less likely to receive this particular vaccine. Children’s household wealth status had shown a strong association with HBV vaccination and the estimated AOR values suggested that those children from the richest households were 88% more likely than the poorest children and even the children from the middle class households were 41% more likely than the poorest children to receive the HBV vaccine. Region wise, the likelihoods of HBV vaccination were also different. It was evident that compared to the children from North India, children from West and Central part of India were less likely to receive the vaccine whereas the children from Eastern India were 89% more likely and children from South were 37% more likely to be immunized against hepatitis B. In contrast, children from the Central region were 12% less likely to receive this vaccine.