Bayesian spatial analysis of child vaccination coverage and associated determinants in Tanzania

Abstract

Introduction

Childhood immunisation is one of the most effective public health interventions, significantly reducing morbidity and mortality from vaccine-preventable diseases. Despite global efforts, many low-income and middle-income countries, including Tanzania, face challenges in achieving high vaccination coverage. This study examined the spatial distribution and determinants of full vaccination coverage among children aged 12–23 months in Tanzania.

Methods

A cross-sectional design was applied using 2022 Tanzania Demographic and Health Survey–Malaria Indicator Survey data, including 2143 children aged 12–23 months. Spatial dependence was analysed with Moran’s I and Local Indicators of Spatial Association to detect clustering. A Bayesian spatial hierarchical logistic regression, specified with the Besag-York-Mollié model and fitted via Integrated Nested Laplace Approximation, assessed individual and community-level determinants of vaccination coverage.

Results

The findings showed that only 54.6% of children were fully vaccinated, which is below the national target and the 90% threshold recommended by the World Health Organization (WHO). Maternal education, antenatal care (ANC) attendance and household wealth were significant predictors of full vaccination. Children of mothers with higher education had higher odds of being fully vaccinated (OR=3.78; 95% CI 1.38 to 12.2), and those whose mothers attended four or more ANC visits were more likely to be fully vaccinated (OR=1.24; 95% CI 0.97 to 1.59). Children from middle-income households also had increased odds (OR=1.39; 95% CI 1.04 to 1.85). Maternal age and place of delivery were not statistically significant. Spatial analysis revealed significant geographic clustering of vaccination coverage (Global Moran’s I=0.229, p=0.028), with high-coverage clusters in Kusini Unguja, Kigoma, Mara and Njombe, and low-coverage clusters in Songwe, Rukwa and Shinyanga.

Conclusions

The study highlights the importance of enhancing maternal education, strengthening ANC counselling, reducing financial and access barriers and implementing region-specific interventions to improve childhood immunisation coverage and reduce inequalities across Tanzania.

WHAT IS ALREADY KNOWN ON THIS TOPIC

• Vaccination saves millions of lives each year, yet coverage remains uneven in sub-Saharan Africa. In Tanzania, only 53% of children were fully vaccinated in 2022, with disparities linked to poverty, geography and maternal factors. Prior studies rarely accounted for spatial variation.

WHAT THIS STUDY ADDS

• Using Bayesian spatial hierarchical logistic regression with the Besag-York-Mollié model, this study identifies maternal education, antenatal care and wealth as key predictors of child vaccination in Tanzania. It also reveals clear geographic clustering of coverage across regions.

HOW THIS STUDY MIGHT AFFECT RESEARCH, PRACTICE OR POLICY

• The findings provide policymakers with evidence to design targeted, region-specific interventions and strengthen maternal education to improve immunisation uptake.

Introduction

Vaccination is widely recognised as one of the most effective public health interventions, significantly reducing morbidity and mortality from vaccine-preventable diseases.1 Through stimulating the immune system to produce protective antibodies, vaccines safeguard against serious infections such as measles, polio, pneumonia and tetanus.2 Childhood immunisation alone prevents more than four million deaths annually and is projected to save approximately 5.2 million lives each year between 2021 and 2030.3 Despite these benefits, disparities in vaccine coverage persist, leaving marginalised and vulnerable populations at higher risk of preventable diseases.4 Globally, nearly 20 million infants miss essential vaccinations each year, highlighting persistent gaps in immunisation efforts.5

To address these challenges, numerous global initiatives have been launched to promote universal immunisation. The Expanded Programme on Immunisation, introduced by the WHO in 1974, laid the foundation for routine vaccination programmes worldwide.6 7 Additionally, organisations such as Global Alliance for Vaccines and Immunisations, and partnerships like WHO-UNICEF immunisation strategies have been instrumental in financing and supporting vaccine delivery in low-income and middle-income countries.8 More recently, the Immunisation Agenda 2030 has set ambitious targets to reduce vaccine-preventable diseases by ensuring that every individual benefits from the full spectrum of vaccines throughout life.9 These efforts have significantly expanded access to vaccines and reduced mortality from childhood illnesses in many countries.

However, these global efforts have faced new challenges in recent years. Particularly, the COVID-19 pandemic disrupted healthcare services worldwide and widened immunity gaps.10 As a result, the number of zero-dose children, those who have never received any vaccine, increased by 40% to 14.5 million by 2023.10 11 Current global coverage for the third dose of the diphtheria, tetanus and pertussis (DTP3) vaccine stands at 84%, while first-dose measles coverage is at 83%, both below pre-pandemic levels.5 Although the introduction of vaccines such as the human papillomavirus vaccine has improved coverage in some regions, persistent disparities remain, especially in low-income countries where weak healthcare infrastructure and logistical barriers hinder timely vaccine delivery.1

These global trends are particularly evident in sub-Saharan Africa, where the burden of under-5 mortality remains disproportionately high. In 2022, the region accounted for 57% of global under-5 deaths despite representing only 30% of the world’s live births.1 Vaccination coverage continues to lag behind global averages, with DTP3 at 76% and first-dose measles coverage at 67%, compared with 83% globally.5 Approximately eight million children in the region remain unvaccinated due to weak healthcare systems, geographical isolation and the lingering effects of the COVID-19 pandemic.8 Nigeria and Ethiopia alone contribute over three million unvaccinated children, underscoring systemic challenges in delivering essential immunisation services.12

Against this regional backdrop, Tanzania illustrates both progress and persistent challenges in achieving universal vaccination coverage. The country has maintained a comprehensive routine immunisation schedule for children under 2 years, which includes the BCG vaccine at birth or during the first healthcare visit; the oral polio vaccine (OPV) at birth, 6, 10 and 14 weeks; the inactivated polio vaccine at 14 weeks; and additional vaccines such as rotavirus, pneumococcal conjugate (PCV13) and pentavalent vaccines at 6, 10 and 14 weeks. The measles-rubella (MR) vaccine is also administered at 9 and 18 months.13

Although child immunisation services are available nationwide in Tanzania, recent data reveal a concerning decline in vaccination coverage. A key measure is the proportion of children who receive all ‘basic’ antigens defined as BCG, three doses each of polio and DPT-containing vaccines, and one measles-containing dose. Among children aged 12–23 months, only 53 percent are fully vaccinated: 91% received BCG, 90% received three DPT doses, 59% received three OPV doses, and 87% received MR.13 Even more worrying, only 49% received all basic vaccines by their first birthday. While coverage remained relatively stable between 1991–1992 (71%) and 2015–2016 (75%), it declined sharply to 53% in 2022. Over the same period, the proportion of children who received no vaccinations fluctuated between 2% and 5%.13

This downward trend is particularly striking given the Tanzanian government’s deliberate efforts to expand access, including exempting vaccination fees and integrating immunisation into other health services. Yet, inequities persist. Children from low-income households and rural areas continue to face barriers such as limited access to healthcare facilities, transportation challenges and financial constraints.14–16 Similarly, nomadic and remote populations record lower vaccination uptake due to logistical difficulties in healthcare delivery.14

Many studies have highlighted similar inequities at both global and regional levels. Factors influencing child vaccination coverage have been widely documented, including socioeconomic status, geographic access and maternal characteristics.14–19 However, most of these analyses rely on conventional statistical methods such as logistic or Poisson regression, which do not adequately capture spatial variation. Moreover, in Tanzania, much of the existing research has focused on specific regions, age groups or ethnic populations.20–24

Recent advances in spatial analytical techniques offer new opportunities to detect geographic clusters, map vaccination gaps and explore their determinants. Although a few studies have examined spatial variation in child vaccination,17 25–27 none have applied flexible structured additive regression models capable of simultaneously estimating regional spatial effects while controlling for individual and household-level covariates in Tanzania. To address this gap, the present study examines the determinants of full vaccination among children aged 12–23 months in Tanzania while accounting for socio-demographic, economic and health-related factors. It incorporates Bayesian spatial hierarchical logistic regression with the Besag-York-Mollié (BYM) specification to explore regional disparities using 2022 Tanzania Demographic and Health Survey (TDHS) data. The findings are expected to guide efforts to increase vaccination coverage, reduce inequities and contribute to the achievement of Sustainable Development Goal 3 on ensuring healthy lives and promoting well-being for all ages by 2030.28

Methods

Study design and data sources

This study used cross-sectional data from the 2022 TDHS and Malaria Indicator Survey (TDHS-MIS).29 The TDHS-MIS, carried out periodically since the 1990s, provides nationally representative insights into a wide range of health indicators. The 2022 round gathered detailed information on maternal and child health, nutrition, and particularly on childhood vaccination coverage.

Sample design and study population

The study focused on children aged 12–23 months, as this age group represents the critical period for evaluating full vaccination coverage.2 13 The sampling procedure followed that of the 2022 TDHS-MIS, which used a stratified two-stage design. In the first stage, 629 clusters were chosen, containing 211 from urban areas and 418 from rural areas. In the second stage, 26 households were systematically selected from each cluster, resulting in a total of 16 354 households.29 After applying inclusion criteria and data cleaning, the final analytical sample entailed 2143 children aged 12–23 months.

Study variables

Outcome variable

The main outcome variable is full vaccination coverage, defined as a binary variable ( $1 = fully vaccinated$ , $0 = not fully vaccinated$ ) indicating whether the child has completed all basic vaccinations by the age of 12–23 months. Full basic vaccination, according to the WHO recommendation, includes one dose of BCG vaccine, three doses of DPT-HepB-Hib, three doses of polio vaccine and one dose of measles vaccine.

Therefore:

$${\displaystyle {\displaystyle \mathrm{f}\mathrm{u}\mathrm{l}\mathrm{l}\mathrm{v}\mathrm{a}\mathrm{c}\mathrm{c}\mathrm{i}\mathrm{n}\mathrm{a}\mathrm{t}\mathrm{i}\mathrm{o}\mathrm{n}\mathrm{c}\mathrm{o}\mathrm{v}\mathrm{e}\mathrm{r}\mathrm{a}\mathrm{g}\mathrm{e}=\{\begin{array}{l}1\mathrm{i}\mathrm{f}\mathrm{c}\mathrm{h}\mathrm{i}\mathrm{l}\mathrm{d}i\mathrm{i}\mathrm{s}\mathrm{f}\mathrm{u}\mathrm{l}\mathrm{l}\mathrm{y}\mathrm{v}\mathrm{a}\mathrm{c}\mathrm{c}\mathrm{i}\mathrm{n}\mathrm{a}\mathrm{t}\mathrm{e}\mathrm{d}\\ 0\mathrm{i}\mathrm{f}\mathrm{c}\mathrm{h}\mathrm{i}\mathrm{l}\mathrm{d}i\mathrm{n}\mathrm{o}\mathrm{t}\mathrm{f}\mathrm{u}\mathrm{l}\mathrm{l}\mathrm{y}\mathrm{v}\mathrm{a}\mathrm{c}\mathrm{c}\mathrm{i}\mathrm{n}\mathrm{a}\mathrm{t}\mathrm{e}\mathrm{d}\end{array}}}$$

Explanatory variables

Explanatory variables for this study were identified from previous literature17 18 30–33 and organised into three categories. Sociodemographic variables included the child’s sex, place of residence, zone of residence, childbirth order, maternal age, maternal education and sex of the household head. Economic variables were represented by the household wealth index, parental employment status and exposure to mass media. Health-related variables comprised the number of antenatal care (ANC) visits, place of delivery and postnatal care visits. The measurement of these variables is reported in the online supplemental table 1.

Statistical analyses

Data were processed and analysed using R software (V.4.5.0). The analytical sample comprised 2143 children aged 12–23 months who were alive at the time of the survey. All included children had complete vaccination records for BCG, pentavalent, polio, rotavirus and measles vaccines, as well as complete information for all covariates considered in the analysis (maternal age, maternal education, household wealth, parity, place of residence and region). Consequently, no children were excluded due to missing data, minimising the risk of selection bias. Detailed information on the exact month or quarter of vaccination or survey administration was not available in the 2022 TDHS-MIS dataset, so temporal variation within the year could not be assessed. Geographic location was assigned based on the child’s region of residence at the time of the survey, rather than the location where vaccination services were received.

Sampling weights, clusters and strata were incorporated to account for the complex survey design. Weighted descriptive statistics and visualisations were generated, followed by univariate logistic regression to assess crude associations. Variables with p≤0.20 in the univariate analysis, or those supported by previous literature, were considered for inclusion in the multivariable models. Importantly, all covariates with strong theoretical or empirical support including maternal education, household wealth and ANC visits were included in the multivariable Bayesian spatial hierarchical model regardless of their univariate p values. This strategy accommodates correlated predictors and mitigates potential bias or instability in model estimates. Reference categories were chosen to represent the most disadvantaged or baseline groups (eg, no maternal education, poorest wealth quintile, no ANC visits), providing a meaningful benchmark for interpreting ORs. This approach prioritises policy relevance and interpretability over category frequency, allowing clear comparisons across socio-demographic and health factors.

Global spatial autocorrelation was assessed using Moran’s I statistic, calculated as:

$${\displaystyle {\displaystyle {\displaystyle I=\frac{\sum { }_{i=1}^N\sum _j{w}_{ij}\left(x_i-\stackrel{-}{X}\right)\left(x_j-\stackrel{-}{X}\right)}{\sum _{i=1}{\left(x_i-\overline{X}\right)}^2}}}}$$

where $x_i$ represents the vaccination rate in region $i$ , $\stackrel{-}{X}$ denotes the national mean vaccination rate, $w_{ij}$ are spatial weights between regions $i$ and $j$ , and $N$ is the number of regions. A positive and statistically significant Moran’s I indicates spatial clustering of similar vaccination coverage levels. Local Indicators of Spatial Association (LISA) were subsequently employed to identify regions with significant local clustering, classifying areas as High-High (regions with high vaccination coverage surrounded by high-coverage neighbours), Low-Low (regions with low coverage surrounded by low-coverage neighbours), High-Low (high-outlier regions surrounded by low-coverage areas), or Low-High (low-outlier regions surrounded by high-coverage areas), with statistical significance assessed via 999 Monte Carlo permutations.

For full vaccination coverage, the Bayesian spatial hierarchical logistic regression using the BYM specification was employed to model region-specific effects. The model was expressed as:

$${\displaystyle {\displaystyle {\displaystyle logit\left(p_i\right)={\beta }_0+\sum _{j=1}^P{\beta }_j{X}_{ij}+{\mu }_i+{\gamma }_i}}}$$

where ${ p}_i$ represents the probability of full vaccination for child $i$ , ${\beta }_0$ is the global intercept, $X_{ij}$ denotes the covariate values for child $i$ , and ${\beta }_{j }$ are regression coefficients with Gaussian priors $\left({\beta }_j\sim N\left(0,{\tau }_j^{-1}\right),{\tau }_j=0.001\right)$ representing vague prior information to ensure minimal influence on the posterior estimates. ${\mu }_i\sim CAR\left({\sigma }_{\mu }^2\right)$ are spatially structured random effects following a conditional autoregressive distribution, while ${\gamma }_i\sim N\left(0,{\sigma }_{\gamma }^2\right)$ are unstructured random effects capturing region-specific heterogeneity. Penalised complexity (PC) priors were specified for both components to stabilise estimation $\left(P\left({\sigma }_{\mu }>1\right)=0.01;P\left({\sigma }_{\gamma }>0.5\right)=0.01\right)$ . Posterior distributions were approximated using Integrated Nested Laplace Approximation (INLA), with posterior means and 95% credible intervals reported. Posterior spatial effects were mapped as $exp({\mu }_i+{\gamma }_i)$ to visualise regions with higher or lower-than-expected vaccination coverage after adjusting for covariates.

Results

Descriptive characteristics of study participants and vaccination coverage

Table 1 presents the characteristics of 2143 children aged 12–23 months, of whom 54.6% were fully vaccinated. Vaccination coverage was slightly higher in urban areas (56.4%) than rural areas (53.9%) and varied across regions, ranging from nearly 70% in Zanzibar to 42% in the Southern zone. Among children by birth order, 57.3% of first-borns were fully vaccinated compared with later-born children. Coverage also varied by maternal education, from 41.8% among children of mothers with no schooling to 75% among those whose mothers had higher education, and by maternal age, ranging from 50% for mothers aged 15–19 years to 56.4% for mothers aged 35–49 years. Children from wealthier households (up to 60%) and those whose mothers had media exposure (57.6%) showed higher vaccination coverage. Among health-related factors, 56.3% of children born in health facilities and 57.7% of those whose mothers attended four or more antenatal visits were fully vaccinated. Vaccination coverage was slightly higher among males (55.5%) than females (53.8%) and slightly higher in male-headed households (54.9%) compared with female-headed households (53.6%).

Table 1.

Distribution of child vaccination coverage by key factors

Variables	Vaccination status
	Fully vaccinated	Not fully vaccinated
	n (%)	n (%)
Sociodemographic characteristics
Sex of the child
Male	607 (54.7)	503 (45.3)
Female	563 (54.5)	470 (45.5)
Place of residence
Urban	332 (56.4)	257 (43.6)
Rural	838 (53.9)	716 (46.1)
Child birth order
1st	275 (57.3)	205 (42.7)
2nd	230 (51.5)	217 (48.5)
3rd	204 (55.3)	165 (44.7)
4th and above	461 (54.4)	386 (45.6)
Maternal education
No education	191 (41.8)	266 (58.2)
Primary	592 (54.9)	487 (45.1)
Secondary	372 (63.5)	215 (36.6)
Higher	15 (75.0)	5 (25.0)
Maternal age
15–19 years	69 (50.0)	69 (50.0)
20–34 years	827 (54.4)	692 (45.6)
35–49 years	274 (56.4)	212 (43.6)
Household head sex
Male	930 (54.6)	772 (45.4)
Female	240 (54.4)	201 (45.6)
Zone of residence
Western	103 (51.5)	97 (48.5)
Northern	99 (55.9)	78 (44.1)
Central	107 (58.2)	77 (41.8)
Southern highlands	104 (63.0)	61 (37.0)
Southern	35 (42.2)	48 (57.8)
South west highlands	123 (45.6)	147 (54.4)
Lake	274 (51.0)	263 (49)
Eastern	101 (49.0)	105 (51.0)
Zanzibar	224 (69.8)	97 (30.2)
Economic characteristics
Wealth Index
Poorest	205 (43.2)	270 (56.8)
Poorer	236 (54.4)	198 (45.6)
Middle	242 (58.3)	173 (41.7)
Richer	252 (60.4)	165 (39.6)
Richest	235 (58.5)	167 (41.5)
Exposure to mass media
Yes	787 (57.6)	579 (42.4)
No	383 (49.3)	394 (50.7)
Mothers working status
Yes	723 (55.4)	583 (44.6)
No	447 (53.4)	390 (46.6)
Health-related characteristics
Place of delivery
Home	191 (47.3)	213 (52.7)
Health facility	979 (56.3)	760 (43.7)
Postnatal care visits
1	998 (54.4)	838 (45.6)
2	162 (56.4)	125 (43.6)
3	10 (50.0)	10 (50.0)
Number of antenatal care visits
None	169 (49.3)	174 (50.7)
1–3 visits	233 (49.7)	236 (50.3)
4 and above visits	768 (57.7)	563 (42.3)

Spatial distribution and clustering of child vaccination across regions in Tanzania

The Global Moran’s I statistic was computed to assess spatial autocorrelation in full vaccination coverage among children aged 12–23 months across regions in Tanzania. The analysis yielded a Moran’s I value of 0.229, with an expected value under the null hypothesis of spatial randomness of –0.033, a variance of 0.019, a standard deviate of 1.906, and a p value of 0.028, indicating statistical significance at the 5% level (see online supplemental table 2). These results suggest moderate positive spatial autocorrelation, meaning that regions with similar vaccination coverage tend to be geographically clustered rather than randomly distributed, reflecting regional influences on vaccination uptake. To further illustrate the spatial pattern, the Moran scatter plot provides a visual representation of this clustering.

The Moran scatter plot in figure 1 shows the relationship between regional full vaccination coverage (x-axis) and the spatially lagged values (y-axis), representing the average coverage in neighbouring regions. The upward-sloping regression line confirms the positive spatial relationship, consistent with the Moran’s I result. Regions in the upper right and lower left quadrants indicate positive spatial autocorrelation, where high-coverage regions are surrounded by high-coverage neighbours and low-coverage regions are adjacent to low-coverage areas. A notable outlier, Mjini Magharibi, lies in the upper-right quadrant, showing exceptionally high vaccination coverage relative to neighbouring regions. Despite this influence, the overall positive spatial autocorrelation indicated by Moran’s I (I=0.229, p=0.028) remains robust across most regions, demonstrating a statistically significant clustering of similar vaccination coverage levels nationwide.

Figure 1.

Moran scatter plot of regional child vaccination coverage.

Building on these global patterns, the LISA cluster map (figure 2), based on Local Moran’s I statistics, revealed significant local spatial autocorrelation in full vaccination coverage. A statistically significant High-High cluster was detected in Kusini Unguja and Mjini Magharibi Pemba, indicating consistently high coverage in these regions and their neighbours. Conversely, a low-low cluster was identified in Songwe, showing a concentration of low vaccination coverage. Kigoma emerged as a high-low outlier, characterised by relatively high coverage surrounded by regions with lower coverage. Regions not classified in these categories were considered not significant, reflecting no detectable local spatial autocorrelation. These findings highlight specific geographic areas where vaccination uptake is notably higher or lower than surrounding regions, suggesting opportunities for targeted, spatially informed public health interventions.

Figure 2.

LISA cluster map of full child vaccination coverage across Tanzania region. LISA, Local Indicators of Spatial Association.

Factor associated with full vaccination coverage

Univariate logistic regression analysis was first conducted to examine the crude associations between each explanatory variable and full child vaccination coverage. Variables that showed significant associations in the univariate analysis, along with those supported by previous literature, were subsequently included in the Bayesian hierarchical model. The results presented in table 2 indicate that maternal education was strongly associated with vaccination: children of mothers with primary, secondary and higher education had higher odds of being fully vaccinated compared with those whose mothers had no education, with ORs of 1.54 (95% CI 1.22 to 1.96), 1.87 (95% CI 1.38 to 2.54) and 3.78 (95% CI 1.38 to 12.2), respectively. The higher education category included only 20 children, producing wide credible intervals in the Bayesian hierarchical model estimates, reflecting greater uncertainty in the effect size for this group. Maternal age, place of delivery and mass media exposure did not show statistically significant associations. Attendance at four or more ANC visits showed higher odds of full vaccination (OR=1.24, 95% CI 0.97 to 1.59), although the association was not statistically significant at the 5% level. Wealth index showed higher odds of full vaccination among children from poorer and middle-income households compared with the poorest, with ORs of 1.33 (95% CI 1.01 to 1.75) and 1.39 (95% CI 1.04 to 1.85), respectively. Zone of residence also influenced coverage: children in the Southern and South West Highlands zones had lower odds of full vaccination (OR=0.53, 95% CI 0.31 to 0.91; OR=0.66, 95% CI 0.45 to 0.96), respectively, while Zanzibar showed a higher likelihood (OR=1.49, 95% CI 1.00 to 2.22) compared with the Western zone.

Table 2.

Results of the Bayesian hierarchical model for the factors associated with child vaccination coverage

Variable	OR (95% CI)	P value
Maternal education
No education (Ref)	Ref	Ref
Primary	1.54 (1.22 to 1.96)	<0.001
Secondary	1.87 (1.38 to 2.54)	<0.001
Higher	3.78 (1.38 to 12.2)	0.015
Maternal age
15–19 (Ref)	Ref	Ref
20–34	1.04 (0.72 to 1.48)	0.849
35–49	1.28 (0.86 to 1.90)	0.222
Place of delivery
Home (Ref)	Ref	Ref
Health facility	1.16 (0.91 to 1.47)	0.241
Number of ANC visits
None (Ref)	Ref	Ref
1–3 visits	1.00 (0.75 to 1.33)	0.997
4+ visits	1.24 (0.97 to 1.59)	0.088
Wealth index
Poorest (Ref)	Ref	Ref
Poorer	1.33 (1.01 to 1.75)	0.041
Middle	1.39 (1.04 to 1.85)	0.027
Richer	1.33 (0.97 to 1.82)	0.076
Richest	1.03 (0.73 to 1.45)	0.885
Mass media exposure
No (Ref)	Ref	Ref
Yes	1.10 (0.90 to 1.36)	0.347
Zone of residence
Western (Ref)	Ref	Ref
Northern	0.97 (0.63 to 1.48)	0.879
Central	1.12 (0.74 to 1.70)	0.596
Southern highlands	1.15 (0.74 to 1.79)	0.532
Southern	0.53 (0.31 to 0.91)	0.02
South west highlands	0.66 (0.45 to 0.96)	0.032
Lake	0.83 (0.59 to 1.17)	0.287
Eastern	0.67 (0.44 to 1.02)	0.06
Zanzibar	1.49 (1.00 to 2.22)	0.053

ANC, antenatal care.

The results of the Bayesian hierarchical spatial model revealed substantial geographic disparities in full vaccination coverage among children aged 12–23 months in Tanzania. As shown in figure 3, the posterior spatial effects highlight regional deviations in vaccination coverage that remain after adjusting for sociodemographic, economic and health service-related covariates.

Figure 3.

Posterior spatial effects of full vaccination coverage among children aged 12–23 months in Tanzania. Positive values (red) indicate regions with above-average coverage after adjusting for model covariates, while negative values (blue) indicate lower-than-expected coverage.

Regions such as Kilimanjaro, Mara, Njombe, Singida and Kigoma display positive spatial effects. This means that children living in these regions tend to have higher vaccination coverage than would be expected based only on their individual and household characteristics. These positive effects reflect favourable local conditions, such as stronger implementation of immunisation programmes, better functioning health services or higher levels of community awareness and support for vaccination. In contrast, regions such as Rukwa and Shinyanga show negative spatial effects, indicating lower vaccination coverage than expected after adjustment for observed factors. This suggests the presence of additional challenges in these areas, like difficulties in accessing health facilities, transportation barriers, health system limitations, or sociocultural factors that discourage vaccine uptake.

The spatial effects observed for Kigoma are consistent with the earlier LISA cluster analysis, where the region appeared as a high-low outlier, meaning it had relatively high vaccination coverage compared with neighbouring regions. The Bayesian spatial model confirms this pattern by showing a positive spatial effect for Kigoma, suggesting that the region benefits from favourable contextual or programme-related factors that support vaccination beyond individual and household characteristics.

These findings are in line with the results from the Bayesian logistic regression model, which showed that children from wealthier households and those whose mothers have higher levels of education are more likely to be fully vaccinated. However, the continued presence of regional differences after accounting for these factors emphasises that geographic location itself plays an important role in shaping vaccination outcomes.

Generally, the results suggest that improving vaccination coverage in Tanzania requires a two-pronged approach: addressing individual-level and household-level factors while also implementing region-specific strategies that respond to local conditions. Targeted interventions that consider regional contexts are therefore essential for reducing persistent vaccination inequalities across the country.

Discussion

This study used data from the 2022 TDHS-MIS to examine the spatial distribution and determinants of full vaccination coverage among children aged 12–23 months in Tanzania, applying a Bayesian spatial hierarchical logistic regression with the BYM specification and implemented via INLA. By integrating spatial effects and posterior estimates, combined with spatial autocorrelation techniques such as Global Moran’s I and LISA, the study identified significant regional disparities and key socio-demographic, economic and health service-related factors associated with childhood immunisation.

The descriptive findings revealed that only 54.6% of children aged 12–23 months were fully vaccinated, well below Tanzania’s national immunisation target and the 90% coverage threshold recommended by the WHO. This underperformance underscores the urgent need to strengthen national immunisation programmes and improve outreach strategies. Low coverage reflects persistent barriers such as weak health infrastructure, limited access and low maternal awareness, which may hinder timely and complete vaccination. Comparable trends in other East African countries, such as Ethiopia and Kenya,22 suggest that regional collaboration and knowledge-sharing could inform effective interventions.

Maternal education emerged as a strong predictor of child vaccination. Children of mothers with secondary or higher education had significantly increased odds of full vaccination compared with those whose mothers had no education. The implication is that improving female education and health literacy could have a long-term effect on childhood immunisation rates. Maternal literacy enhances health-seeking behaviour, strengthens understanding of immunisation messages and increases adherence to vaccination schedules, as observed in Nigeria31 and broader sub-Saharan Africa.34 Health promotion programmes should therefore integrate educational interventions targeting mothers, especially in regions with low coverage.

ANC visits were positively associated with full vaccination, with mothers attending four or more ANC visits more likely to vaccinate their children. This finding implies that strengthening ANC services and ensuring comprehensive counselling on child immunisation during pregnancy could improve vaccine uptake. Health systems should emphasise ANC as a platform for promoting child health interventions, echoing findings from Malawi and Ethiopia.35

Household wealth had a modest but positive influence on vaccination coverage. Children from middle or wealthier households were more likely to be fully vaccinated, highlighting that financial resources can reduce logistical barriers, including transport costs and service fees. Programmes targeting poorer households with subsidies, outreach campaigns or mobile vaccination services could help mitigate inequities, as noted in Burkina Faso and Nigeria.36 However, the plateau in the richest quintile suggests that private healthcare use, which was not captured in national records, could influence observed coverage patterns.

Interestingly, maternal age and place of delivery were not statistically significant predictors after adjusting for other variables. This indicates that, in the Tanzanian context, these factors are less influential when broader service quality and community-level determinants are accounted for. Policies should therefore focus on systemic improvements rather than solely targeting maternal age or delivery location, contrasting with findings in other settings.37

Spatial analysis revealed significant regional clustering in vaccination coverage. The Global Moran’s I value of 0.229 (p=0.028) confirmed spatial autocorrelation, indicating that high and low coverage areas tend to be geographically concentrated. This suggests that interventions should be geographically targeted, prioritising low-coverage regions for intensified outreach and resource allocation. Similar patterns were observed in Mozambique and Ethiopia,27 where uneven health infrastructure and programme reach contributed to disparities.

LISA analysis identified High-High clusters in Kusini Unguja and Mjini Magharibi Pemba and low-low clusters in Songwe, pinpointing areas of programmatic success and urgent need, respectively. The implications are clear: regions with high coverage could serve as models for best practices, while low-coverage areas require tailored strategies, such as strengthening local health system capacity, community engagement and vaccination campaigns.38

The Bayesian spatial model further highlighted that Kigoma, Mara and Njombe exhibited above-average spatial effects, implying positive unmeasured contextual influences, due to successful immunisation campaigns and non-government organisation (NGO) support. Conversely, Rukwa and Shinyanga had significantly negative spatial effects, indicating enduring structural barriers such as difficult terrain, vaccine hesitancy and limited-service delivery. These results reinforce the need for geographically informed policy decisions and targeted resource allocation to address inequities, consistent with findings in Nigeria where marginalised areas faced structural health barriers.39

It is important to note that while the LISA analysis identified Kigoma as a High–Low outlier based on raw vaccination coverage, the Bayesian spatial model also revealed a positive spatial effect for the region after adjusting for individual-level and household-level factors. This consistency suggests that Kigoma’s relatively high vaccination coverage is not explained solely by favourable sociodemographic characteristics but is also supported by positive regional contextual factors, such as effective immunisation service delivery, supportive local health system structures or strong community engagement. The consistency of the findings between the descriptive LISA results and the adjusted Bayesian estimates strengthens the evidence that geographic context plays an important role in shaping vaccination outcomes in the region.

Strengths and limitations

The current study has several strengths. It used advanced Bayesian spatial hierarchical modelling with the BYM specification, accounted for spatial dependence to reveal geographic clustering, and used nationally representative TDHS-MIS 2022 data, ensuring generalisability. However, there are several limitations. Cultural factors were not captured, and the analysis considered full vaccination rather than individual vaccines. Temporal heterogeneity within 2022 could not be assessed, as detailed information on the month or quarter of vaccination or survey administration was not available, which may have obscured potential variations in vaccination coverage related to COVID-19 service disruptions and seasonal trends. Geographic location was defined based on the child’s region of residence at the time of the survey rather than the location where vaccination was received, which could result in misclassification if families moved across regions. The small number of children whose mothers had higher education may have resulted in imprecise estimates and wide CIs for this category. Residual confounding from unmeasured factors, such as cultural practices or local service delivery variations, may influence the observed associations and limit causal interpretation. Finally, the use of secondary TDHS-MIS data limited the ability to measure additional factors that could have been captured through primary data collection, such as more detailed behavioural, cultural or service delivery determinants. Future studies could address these gaps by integrating cultural aspects, examining specific vaccines, incorporating temporal trends when data allow, and performing sensitivity analyses to verify model stability and robustness, and collecting primary data to capture additional relevant determinants.

Conclusions

The current study explores the spatial distribution and determinants of full vaccination coverage among children aged 12–23 months in Tanzania using data from the 2022 TDHS-MIS. Significant determinants of full vaccination were identified, including maternal education, ANC attendance and household wealth, while maternal age and place of delivery were not statistically significant after adjustment. The spatial analysis revealed notable geographic disparities, with high-coverage areas in Kusini Unguja, Kigoma, Mara and Njombe, and low-coverage areas in Songwe, Rukwa and Shinyanga.

The findings indicate that children of educated mothers, from wealthier households, and whose mothers attended four or more ANC visits had higher odds of being fully vaccinated. Regional and local factors also played an important role, as shown by clustering patterns detected through Global Moran’s I and LISA analyses. Based on these results, policy recommendations include enhancing maternal education and health literacy programmes, strengthening ANC services with vaccine counselling, addressing financial and access barriers for disadvantaged households, and implementing region-specific interventions to target low-coverage areas. Continued monitoring and context-specific strategies are essential to improve childhood immunisation coverage and reduce inequalities across Tanzania.

Data availability statement

Data are available in a public, open access repository. The dataset analysed in this study is publicly available from the Tanzania Demographic and Health Survey (TDHS) Program. Researchers can request access through the DHS Program website: https://dhsprogram.com.

Ethics statements

Patient consent for publication

Not applicable.

Ethics approval

This study used secondary data; therefore, obtaining ethical approval was not necessary. However, permission to access and use the data was requested from the DHS Program. The 2022 TDHS-MIS received ethical clearance from various national review bodies, including the National Institute for Medical Research and the Zanzibar Medical Research Ethical Committee. Furthermore, to ensure transparent reporting, the study complies with the Strengthening the Reporting of Observational Studies in Epidemiology checklist, which is included in online supplemental table 3.