Modelling zero dose prevalence in Kenya using geostatistical and GeoAI techniques

Background: Globally, an estimated 14.5 million children remain unvaccinated, with the largest proportion residing in SSA. In many SSA countries, vaccine access remains highly unequal, with low uptake frequently driven by conflict, fragility, population displacement, and ongoing humanitarian crises. Even in countries reporting high national full-immunization coverage, such as Rwanda (95.5%) and Kenya (80%), clusters of under-immunized and ZD children persist in marginalized, underserved, and hard-to-reach settings.

Efforts to improve vaccination uptake in SSA are further constrained by limited data availability, compromised data quality and insufficient use of advanced spatial and statistical methods needed to uncover small, high-risk pockets of ZD children. Moreover, no synthesized review of the spatial datasets and methods used to estimate vaccine coverage at high resolution, that can guide ZD modelling studies, exists. Most studies rely solely on survey data, overlooking the complementary strengths of routine data, largely due to capacity limitations in modelling routine data or jointly analysing it alongside survey data. Such modelling efforts are particularly important for identifying and addressing immunization gaps in marginalized and underserved areas, where ZD children are most likely to be concentrated.

Methods: The study will begin with a scoping review of spatial datasets and the spatial or spatial-statistical methods used to map ZD prevalence in LLMICs. This review will also identify key gaps and limitations in current spatial data and methods. Insights from the review will then inform covariate selection and the development of a novel joint modelling framework for estimating ZD prevalence using both routine and survey data. To identify areas of heightened vulnerability to ZD, a ZDVI will be constructed using model-based geostatistical methods and a set of relevant covariates – including factors reflecting marginalization and limited healthcare access (e.g., travel time to health facilities, conflict exposure). Covariate weights will be derived through Analytic Hierarchy Process (AHP) multi-criteria decision analysis, administered via an online questionnaire distributed to immunization experts, healthcare workers, and other stakeholders. Finally, the study will undertake a focused analysis of selected marginalized, nomadic counties to examine the effects of population migration on vaccination uptake in these settings.

Expected results: This research will generate key insights into the spatial datasets and analytical methods used to identify ZD children – particularly at high spatial resolution, which is critical given that ZD children often occur in small clusters. These findings will establish a strong foundation for future studies investigating ZD prevalence in LLMIC settings. Joint modelling of ZD prevalence using both routine and survey data is expected to produce robust, high-quality estimates for Kenya, while also offering a transferable framework that enables other contexts to make more effective use of routine data and highlighting the strongest predictors of ZD status. Further, the ZDVI will synthesize key risk factors associated with ZD, allowing identification of areas with the highest vulnerability. Finally, by examining vaccination uptake in nomadic counties, the study will improve understanding of how migration affects immunization coverage among nomadic and hard-to-reach communities.